Computational methods for metastasis detection in lymph nodes and characterization of the metastasis-free lymph node microarchitecture: A systematic-narrative hybrid review

Background

Histological examination of tumor draining lymph nodes (LNs) plays a vital role in cancer staging and prognostication. However, as soon as a LN is classed as metastasis-free, no further investigation will be performed and thus, potentially clinically relevant information detectable in tumor-free LNs is currently not captured.

Objective

To systematically study and critically assess methods for the analysis of digitized histological LN images described in published research.

Methods

A systematic search was conducted in several public databases up to December 2023 using relevant search terms. Studies using brightfield light microscopy images of hematoxylin and eosin or immunohistochemically stained LN tissue sections aiming to detect and/or segment LNs, their compartments or metastatic tumor using artificial intelligence (AI) were included. Dataset, AI methodology, cancer type, and study objective were compared between articles.

Results

A total of 7201 articles were collected and 73 articles remained for detailed analyses after article screening. Of the remaining articles, 86% aimed at LN metastasis identification, 8% aimed at LN compartment segmentation, and remaining focused on LN contouring. Furthermore, 78% of articles used patch classification and 22% used pixel segmentation models for analyses. Five out of six studies (83%) of metastasis-free LNs were performed on publicly unavailable datasets, making quantitative article comparison impossible.

Conclusions

Multi-scale models mimicking multiple microscopy zooms show promise for computational LN analysis. Large-scale datasets are needed to establish the clinical relevance of analyzing metastasis-free LN in detail. Further research is needed to identify clinically interpretable metrics for LN compartment characterization.

Development and benchmarking of a Deep Learning-based MRI-guided gross tumor segmentation algorithm for Radiomics analyses in extremity soft tissue sarcomas

BACKGROUND

Volume of interest (VOI) segmentation is a crucial step for Radiomics analyses and radiotherapy (RT) treatment planning. Because it can be time-consuming and subject to inter-observer variability, we developed and tested a Deep Learning-based automatic segmentation (DLBAS) algorithm to reproducibly predict the primary gross tumor as VOI for Radiomics analyses in extremity soft tissue sarcomas (STS).

METHODS

A DLBAS algorithm was trained on a cohort of 157 patients and externally tested on an independent cohort of 87 patients using contrast-enhanced MRI. Manual tumor delineations by a radiation oncologist served as ground truths (GTs). A benchmark study with 20 cases from the test cohort compared the DLBAS predictions against manual VOI segmentations of two residents (ERs) and clinical delineations of two radiation oncologists (ROs). The ROs rated DLBAS predictions regarding their direct applicability.

RESULTS

The DLBAS achieved a median dice similarity coefficient (DSC) of 0.88 against the GTs in the entire test cohort (interquartile range (IQR): 0.11) and a median DSC of 0.89 (IQR 0.07) and 0.82 (IQR 0.10) in comparison to ERs and ROs, respectively. Radiomics feature stability was high with a median intraclass correlation coefficient of 0.97, 0.95 and 0.94 for GTs, ERs, and ROs, respectively. DLBAS predictions were deemed clinically suitable by the two ROs in 35% and 20% of cases, respectively.

CONCLUSION

The results demonstrate that the DLBAS algorithm provides reproducible VOI predictions for radiomics feature extraction. Variability remains regarding direct clinical applicability of predictions for RT treatment planning.

Development of Clinical Radiomics-Based Models to Predict Survival Outcome in Pancreatic Ductal Adenocarcinoma: A Multicenter Retrospective Study

PURPOSE

This multicenter retrospective study aims to identify reliable clinical and radiomic features to build machine learning models that predict progression-free survival (PFS) and overall survival (OS) in pancreatic ductal adenocarcinoma (PDAC) patients.

METHODS

Between 2010 and 2020 pre-treatment contrast-enhanced CT scans of 287 pathology-confirmed PDAC patients from two sites of the Hopital Universitaire de Bruxelles (HUB) and from 47 hospitals within the HUB network were retrospectively analysed. Demographic, clinical, and survival data were also collected. Gross tumour volume (GTV) and non-tumoral pancreas (RPV) were semi-manually segmented and radiomics features were extracted. Patients from two HUB sites comprised the training dataset, while those from the remaining 47 hospitals of the HUB network constituted the testing dataset. A three-step method was used for feature selection. Based on the GradientBoostingSurvivalAnalysis classifier, different machine learning models were trained and tested to predict OS and PFS. Model performances were assessed using the C-index and Kaplan-Meier curves. SHAP analysis was applied to allow for post hoc interpretability.

RESULTS

A total of 107 radiomics features were extracted from each of the GTV and RPV. Fourteen subgroups of features were selected: clinical, GTV, RPV, clinical & GTV, clinical & GTV & RPV, GTV-volume and RPV-volume both for OS and PFS. Subsequently, 14 Gradient Boosting Survival Analysis models were trained and tested. In the testing dataset, the clinical & GTV model demonstrated the highest performance for OS (C-index: 0.72) among all other models, while for PFS, the clinical model exhibited a superior performance (C-index: 0.70).

CONCLUSIONS

An integrated approach, combining clinical and radiomics features, excels in predicting OS, whereas clinical features demonstrate strong performance in PFS prediction.

Deep Learning of Multimodal Ultrasound: Stratifying the Response to Neoadjuvant Chemotherapy in Breast Cancer Before Treatment

BACKGROUND

Not only should resistance to neoadjuvant chemotherapy (NAC) be considered in patients with breast cancer but also the possibility of achieving a pathologic complete response (PCR) after NAC. Our study aims to develop 2 multimodal ultrasound deep learning (DL) models to noninvasively predict resistance and PCR to NAC before treatment.

METHODS

From January 2017 to July 2022, a total of 170 patients with breast cancer were prospectively enrolled. All patients underwent multimodal ultrasound examination (grayscale 2D ultrasound and ultrasound elastography) before NAC. We combined clinicopathological information to develop 2 DL models, DL_Clinical_resistance and DL_Clinical_PCR, for predicting resistance and PCR to NAC, respectively. In addition, these 2 models were combined to stratify the prediction of response to NAC.

RESULTS

In the test cohort, DL_Clinical_resistance had an AUC of 0.911 (95%CI, 0.814-0.979) with a sensitivity of 0.905 (95%CI, 0.765-1.000) and an NPV of 0.882 (95%CI, 0.708-1.000). Meanwhile, DL_Clinical_PCR achieved an AUC of 0.880 (95%CI, 0.751-0.973) and sensitivity and NPV of 0.875 (95%CI, 0.688-1.000) and 0.895 (95%CI, 0.739-1.000), respectively. By combining DL_Clinical_resistance and DL_Clinical_PCR, 37.1% of patients with resistance and 25.7% of patients with PCR were successfully identified by the combined model, suggesting that these patients could benefit by an early change of treatment strategy or by implementing an organ preservation strategy after NAC.

CONCLUSIONS

The proposed DL_Clinical_resistance and DL_Clinical_PCR models and combined strategy have the potential to predict resistance and PCR to NAC before treatment and allow stratified prediction of NAC response.

A clinically relevant computed tomography (CT) radiomics strategy for intracranial rodent brain tumour monitoring

Here, we establish a CT-radiomics based method for application in invasive, orthotopic rodent brain tumour models. Twenty four NOD/SCID mice were implanted with U87R-Luc2 GBM cells and longitudinally imaged via contrast enhanced (CE-CT) imaging. Pyradiomics was employed to extract CT-radiomic features from the tumour-implanted hemisphere and non-tumour-implanted hemisphere of acquired CT-scans. Inter-correlated features were removed (Spearman correlation > 0.85) and remaining features underwent predictive analysis (recursive feature elimination or Boruta algorithm). An area under the curve of the receiver operating characteristic curve was implemented to evaluate radiomic features for their capacity to predict defined outcomes. Firstly, we identified a subset of radiomic features which distinguish the tumour-implanted hemisphere and non- tumour-implanted hemisphere (i.e, tumour presence from normal tissue). Secondly, we successfully translate preclinical CT-radiomic pipelines to GBM patient CT scans (n = 10), identifying similar trends in tumour-specific feature intensities (E.g. 'glszm Zone Entropy'), thereby suggesting a mouse-to-human species conservation (a conservation of radiomic features across species). Thirdly, comparison of features across timepoints identify features which support preclinical tumour detection earlier than is possible by visual assessment of CT scans. This work establishes robust, preclinical CT-radiomic pipelines and describes the application of CE-CT for in-depth orthotopic brain tumour monitoring. Overall we provide evidence for the role of pre-clinical 'discovery' radiomics in the neuro-oncology space.

An Interpretable Radiomics Model Based on Two-Dimensional Shear Wave Elastography for Predicting Symptomatic Post-Hepatectomy Liver Failure in Patients with Hepatocellular Carcinoma

OBJECTIVE

The aim of this study was to develop and validate an interpretable radiomics model based on two-dimensional shear wave elastography (2D-SWE) for symptomatic post-hepatectomy liver failure (PHLF) prediction in patients undergoing liver resection for hepatocellular carcinoma (HCC).

METHODS

A total of 345 consecutive patients were enrolled. A five-fold cross-validation was performed during training, and the models were evaluated in the independent test cohort. A multi-patch radiomics model was established based on the 2D-SWE images for predicting symptomatic PHLF. Clinical features were incorporated into the models to train the clinical-radiomics model. The radiomics model and the clinical-radiomics model were compared with the clinical model comprising clinical variables and other clinical predictive indices, including the model for end-stage liver disease (MELD) score and albumin-bilirubin (ALBI) score. Shapley Additive exPlanations (SHAP) was used for post hoc interpretability of the radiomics model.

RESULTS

The clinical-radiomics model achieved an AUC of 0.867 (95% CI 0.787-0.947) in the five-fold cross-validation, and this score was higher than that of the clinical model (AUC: 0.809; 95% CI: 0.715-0.902) and the radiomics model (AUC: 0.746; 95% CI: 0.681-0.811). The clinical-radiomics model showed an AUC of 0.822 in the test cohort, higher than that of the clinical model (AUC: 0.684, p = 0.007), radiomics model (AUC: 0.784, p = 0.415), MELD score (AUC: 0.529, p < 0.001), and ALBI score (AUC: 0.644, p = 0.016). The SHAP analysis showed that the first-order radiomics features, including first-order maximum 64 × 64, first-order 90th percentile 64 × 64, and first-order 10th percentile 32 × 32, were the most important features for PHLF prediction.

CONCLUSION

An interpretable clinical-radiomics model based on 2D-SWE and clinical variables can help in predicting symptomatic PHLF in HCC.

The clinical importance of the host anti-tumour reaction patterns in regional tumour draining lymph nodes in patients with locally advanced resectable gastric cancer: a systematic review and meta-analysis

BACKGROUND

The status of regional tumour draining lymph nodes (LN) is crucial for prognostic evaluation in gastric cancer (GaC) patients. Changes in lymph node microarchitecture, such as follicular hyperplasia (FH), sinus histiocytosis (SH), or paracortical hyperplasia (PH), may be triggered by the anti-tumour immune response. However, the prognostic value of these changes in GaC patients is unclear.

METHODS

A systematic search in multiple databases was conducted to identify studies on the prognostic value of microarchitecture changes in regional tumour-negative and tumour-positive LNs measured on histopathological slides. Since the number of GaC publications was very limited, the search was subsequently expanded to include junctional and oesophageal cancer (OeC).

RESULTS

A total of 28 articles (17 gastric cancer, 11 oesophageal cancer) met the inclusion criteria, analyzing 26,503 lymph nodes from 3711 GaC and 1912 OeC patients. The studies described eight different types of lymph node microarchitecture changes, categorized into three patterns: hyperplasia (SH, FH, PH), cell-specific infiltration (dendritic cells, T cells, neutrophils, macrophages), and differential gene expression. Meta-analysis of five GaC studies showed a positive association between SH in tumour-negative lymph nodes and better 5-year overall survival. Pooled risk ratios for all LNs showed increased 5-year overall survival for the presence of SH and PH.

CONCLUSIONS

This systematic review suggests that sinus histiocytosis and paracortical hyperplasia in regional tumour-negative lymph nodes may provide additional prognostic information for gastric and oesophageal cancer patients. Further studies are needed to better understand the lymph node reaction patterns and explore their impact of chemotherapy treatment and immunotherapy efficacy.

MSCDA: Multi-level semantic-guided contrast improves unsupervised domain adaptation for breast MRI segmentation in small datasets

Deep learning (DL) applied to breast tissue segmentation in magnetic resonance imaging (MRI) has received increased attention in the last decade, however, the domain shift which arises from different vendors, acquisition protocols, and biological heterogeneity, remains an important but challenging obstacle on the path towards clinical implementation. In this paper, we propose a novel Multi-level Semantic-guided Contrastive Domain Adaptation (MSCDA) framework to address this issue in an unsupervised manner. Our approach incorporates self-training with contrastive learning to align feature representations between domains. In particular, we extend the contrastive loss by incorporating pixel-to-pixel, pixel-to-centroid, and centroid-to-centroid contrasts to better exploit the underlying semantic information of the image at different levels. To resolve the data imbalance problem, we utilize a category-wise cross-domain sampling strategy to sample anchors from target images and build a hybrid memory bank to store samples from source images. We have validated MSCDA with a challenging task of cross-domain breast MRI segmentation between datasets of healthy volunteers and invasive breast cancer patients. Extensive experiments show that MSCDA effectively improves the model's feature alignment capabilities between domains, outperforming state-of-the-art methods. Furthermore, the framework is shown to be label-efficient, achieving good performance with a smaller source dataset. The code is publicly available at https://github.com/ShengKuangCN/MSCDA.

Combining Deep Learning and Handcrafted Radiomics for Classification of Suspicious Lesions on Contrast-enhanced Mammograms

Background Handcrafted radiomics and deep learning (DL) models individually achieve good performance in lesion classification (benign vs malignant) on contrast-enhanced mammography (CEM) images. Purpose To develop a comprehensive machine learning tool able to fully automatically identify, segment, and classify breast lesions on the basis of CEM images in recall patients. Materials and Methods CEM images and clinical data were retrospectively collected between 2013 and 2018 for 1601 recall patients at Maastricht UMC+ and 283 patients at Gustave Roussy Institute for external validation. Lesions with a known status (malignant or benign) were delineated by a research assistant overseen by an expert breast radiologist. Preprocessed low-energy and recombined images were used to train a DL model for automatic lesion identification, segmentation, and classification. A handcrafted radiomics model was also trained to classify both human- and DL-segmented lesions. Sensitivity for identification and the area under the receiver operating characteristic curve (AUC) for classification were compared between individual and combined models at the image and patient levels. Results After the exclusion of patients without suspicious lesions, the total number of patients included in the training, test, and validation data sets were 850 (mean age, 63 years ± 8 [SD]), 212 (62 years ± 8), and 279 (55 years ± 12), respectively. In the external data set, lesion identification sensitivity was 90% and 99% at the image and patient level, respectively, and the mean Dice coefficient was 0.71 and 0.80 at the image and patient level, respectively. Using manual segmentations, the combined DL and handcrafted radiomics classification model achieved the highest AUC (0.88 [95% CI: 0.86, 0.91]) (P < .05 except compared with DL, handcrafted radiomics, and clinical features model, where P = .90). Using DL-generated segmentations, the combined DL and handcrafted radiomics model showed the highest AUC (0.95 [95% CI: 0.94, 0.96]) (P < .05). Conclusion The DL model accurately identified and delineated suspicious lesions on CEM images, and the combined output of the DL and handcrafted radiomics models achieved good diagnostic performance. © RSNA, 2023 Supplemental material is available for this article. See also the editorial by Bahl and Do in this issue.

External validation of 18F-FDG PET-based radiomic models on identification of residual oesophageal cancer after neoadjuvant chemoradiotherapy

OBJECTIVES

Detection of residual oesophageal cancer after neoadjuvant chemoradiotherapy (nCRT) is important to guide treatment decisions regarding standard oesophagectomy or active surveillance. The aim was to validate previously developed 18F-FDG PET-based radiomic models to detect residual local tumour and to repeat model development (i.e. 'model extension') in case of poor generalisability.

METHODS

This was a retrospective cohort study in patients collected from a prospective multicentre study in four Dutch institutes. Patients underwent nCRT followed by oesophagectomy between 2013 and 2019. Outcome was tumour regression grade (TRG) 1 (0% tumour) versus TRG 2-3-4 (≥1% tumour). Scans were acquired according to standardised protocols. Discrimination and calibration were assessed for the published models with optimism-corrected AUCs >0.77. For model extension, the development and external validation cohorts were combined.

RESULTS

Baseline characteristics of the 189 patients included [median age 66 years (interquartile range 60-71), 158/189 male (84%), 40/189 TRG 1 (21%) and 149/189 (79%) TRG 2-3-4] were comparable to the development cohort. The model including cT stage plus the feature 'sum entropy' had best discriminative performance in external validation (AUC 0.64, 95% confidence interval 0.55-0.73), with a calibration slope and intercept of 0.16 and 0.48 respectively. An extended bootstrapped LASSO model yielded an AUC of 0.65 for TRG 2-3-4 detection.

CONCLUSION

The high predictive performance of the published radiomic models could not be replicated. The extended model had moderate discriminative ability. The investigated radiomic models appeared inaccurate to detect local residual oesophageal tumour and cannot be used as an adjunct tool for clinical decision-making in patients.

Towards texture accurate slice interpolation of medical images using PixelMiner

Quantitative image analysis models are used for medical imaging tasks such as registration, classification, object detection, and segmentation. For these models to be capable of making accurate predictions, they need valid and precise information. We propose PixelMiner, a convolution-based deep-learning model for interpolating computed tomography (CT) imaging slices. PixelMiner was designed to produce texture-accurate slice interpolations by trading off pixel accuracy for texture accuracy. PixelMiner was trained on a dataset of 7829 CT scans and validated using an external dataset. We demonstrated the model's effectiveness by using the structural similarity index (SSIM), peak signal to noise ratio (PSNR), and the root mean squared error (RMSE) of extracted texture features. Additionally, we developed and used a new metric, the mean squared mapped feature error (MSMFE). The performance of PixelMiner was compared to four other interpolation methods: (tri-)linear, (tri-)cubic, windowed sinc (WS), and nearest neighbor (NN). PixelMiner produced texture with a significantly lowest average texture error compared to all other methods with a normalized root mean squared error (NRMSE) of 0.11 (p < .01), and the significantly highest reproducibility with a concordance correlation coefficient (CCC) ≥ 0.85 (p < .01). PixelMiner was not only shown to better preserve features but was also validated using an ablation study by removing auto-regression from the model and was shown to improve segmentations on interpolated slices.

From Head and Neck Tumour and Lymph Node Segmentation to Survival Prediction on PET/CT: An End-to-End Framework Featuring Uncertainty, Fairness, and Multi-Region Multi-Modal Radiomics

Automatic delineation and detection of the primary tumour (GTVp) and lymph nodes (GTVn) using PET and CT in head and neck cancer and recurrence-free survival prediction can be useful for diagnosis and patient risk stratification. We used data from nine different centres, with 524 and 359 cases used for training and testing, respectively. We utilised posterior sampling of the weight space in the proposed segmentation model to estimate the uncertainty for false positive reduction. We explored the prognostic potential of radiomics features extracted from the predicted GTVp and GTVn in PET and CT for recurrence-free survival prediction and used SHAP analysis for explainability. We evaluated the bias of models with respect to age, gender, chemotherapy, HPV status, and lesion size. We achieved an aggregate Dice score of 0.774 and 0.760 on the test set for GTVp and GTVn, respectively. We observed a per image false positive reduction of 19.5% and 7.14% using the uncertainty threshold for GTVp and GTVn, respectively. Radiomics features extracted from GTVn in PET and from both GTVp and GTVn in CT are the most prognostic, and our model achieves a C-index of 0.672 on the test set. Our framework incorporates uncertainty estimation, fairness, and explainability, demonstrating the potential for accurate detection and risk stratification.

The application of artificial intelligence in the detection of basal cell carcinoma: A systematic review

Basal cell carcinoma (BCC) is one of the most common types of cancer. The growing incidence worldwide and the need for fast, reliable and less invasive diagnostic techniques make a strong case for the application of different artificial intelligence techniques for detecting and classifying BCC and its subtypes. We report on the current evidence regarding the application of handcrafted and deep radiomics models used for the detection and classification of BCC in dermoscopy, optical coherence tomography and reflectance confocal microscopy. We reviewed all the articles that were published in the last 10 years in PubMed, Web of Science and EMBASE, and we found 15 articles that met the inclusion criteria. We included articles that are original, written in English, focussing on automated BCC detection in our target modalities and published within the last 10 years in the field of dermatology. The outcomes from the selected publications are presented in three categories depending on the imaging modality and to allow for comparison. The majority of articles (n = 12) presented different AI solutions for the detection and/or classification of BCC in dermoscopy images. The rest of the publications presented AI solutions in OCT images (n = 2) and RCM (n = 1). In addition, we provide future directions for the application of these techniques for the detection of BCC. In conclusion, the reviewed publications demonstrate the potential benefit of AI in the detection of BCC in dermoscopy, OCT and RCM.

Outcome prediction in resectable esophageal adenocarcinoma based on clinical variables, radiomics, and ctDNA

423

Background: Despite the advent of precision medicine, prediction of survival outcome of esophageal cancer patients remains a challenge. Here we aim to investigate the value of prediction models integrating multi-signal data including radiomics and circulating tumor DNA (ctDNA) data in addition to clinical data for the prediction of resectable esophageal adenocarcinoma (rEAC) related outcomes. Methods: In total n=111 rEAC patients treated with neoadjuvant chemoradiotherapy (nCRT; n=71) +/- anti-PD-L1 (n=40) were included. Baseline clinical variables (n=9) were based on the SOURCE survival prediction model (van den Boorn et al. JNCCN. 2021). The baseline ctDNA data from plasma was derived from fragmentomic and copy number aberrations (ichorCNA) from shallow whole genome sequencing (<5-fold coverage) and a custom next-generation sequencing panel (n=23 genes). Baseline radiomic original features were extracted by the pyradiomics package from CT-image delineated tumor volumes. An initial redundancy filtering was performed to remove correlating variables (r>0.6). We evaluated the predictive performance of baseline ctDNA and radiomics features on overall survival (OS), progression free survival (PFS), and time to progression (TTP), through fitting Cox-regression models. Four ctDNA features were included in the models: P20-150, ichorCNA, fragment end score and mutation detection. For the radiomics features we performed an additional back- and forward variable selection based on Akaike’s Information Criterion. Using the likelihood ratio test we tested if the model fit changed after adding ctDNA and radiomics features to a model with clinical variables. Results: The addition of radiomics to clinical variables improved model fit for OS (p=0.017). Baseline prediction of OS resulted in a C-index of 0.65 using clinical variables only, 0.65 with ctDNA, 0.68 with radiomics and 0.68 with ctDNA and radiomics combined. For PFS model fit improved after adding radiomics (p=0.020) and ctDNA and radiomics combined (p=0.017). Baseline prediction of PFS resulted in a C-index of 0.64 using clinical variables, 0.65 with ctDNA, 0.67 with radiomics, and 0.68 with ctDNA and radiomics combined. For TTP model fit improved after adding radiomics (p=0.008) and radiomics and ctDNA combined (p=0.002). Baseline prediction of TTP resulted in a C-index of 0.64 with clinical variables, 0.65 with ctDNA, 0.71 with radiomics, and 0.72 with ctDNA and radiomics combined. Based on the cox-regression models using clinical variables and radiomics, risk stratification by splitting the cohort in a high and low risk group was possible for OS, PFS and TTP (p<0.001). Conclusions: Combining clinical variables from SOURCE with radiomics data improved predictions of OS, PFS, and TTP among patients with rEAC. Multi-signal integration of clinical and radiomics variables could potentially be used to identify risk groups.

Deep learning based identification of bone scintigraphies containing metastatic bone disease foci

PURPOSE

Metastatic bone disease (MBD) is the most common form of metastases, most frequently deriving from prostate cancer. MBD is screened with bone scintigraphy (BS), which have high sensitivity but low specificity for the diagnosis of MBD, often requiring further investigations. Deep learning (DL) - a machine learning technique designed to mimic human neuronal interactions- has shown promise in the field of medical imaging analysis for different purposes, including segmentation and classification of lesions. In this study, we aim to develop a DL algorithm that can classify areas of increased uptake on bone scintigraphy scans.

METHODS

We collected 2365 BS from three European medical centres. The model was trained and validated on 1203 and 164 BS scans respectively. Furthermore we evaluated its performance on an external testing set composed of 998 BS scans. We further aimed to enhance the explainability of our developed algorithm, using activation maps. We compared the performance of our algorithm to that of 6 nuclear medicine physicians.

RESULTS

The developed DL based algorithm is able to detect MBD on BSs, with high specificity and sensitivity (0.80 and 0.82 respectively on the external test set), in a shorter time compared to the nuclear medicine physicians (2.5 min for AI and 30 min for nuclear medicine physicians to classify 134 BSs). Further prospective validation is required before the algorithm can be used in the clinic.

Automated detection and delineation of lymph nodes in haematoxylin & eosin stained digitised slides

Treatment of patients with oesophageal and gastric cancer (OeGC) is guided by disease stage, patient performance status and preferences. Lymph node (LN) status is one of the strongest prognostic factors for OeGC patients. However, survival varies between patients with the same disease stage and LN status. We recently showed that LN size from patients with OeGC might also have prognostic value, thus making delineations of LNs essential for size estimation and the extraction of other imaging biomarkers. We hypothesized that a machine learning workflow is able to: (1) find digital H&E stained slides containing LNs, (2) create a scoring system providing degrees of certainty for the results, and (3) delineate LNs in those images. To train and validate the pipeline, we used 1695 H&E slides from the OE02 trial. The dataset was divided into training (80%) and validation (20%). The model was tested on an external dataset of 826 H&E slides from the OE05 trial. U-Net architecture was used to generate prediction maps from which predefined features were extracted. These features were subsequently used to train an XGBoost model to determine if a region truly contained a LN. With our innovative method, the balanced accuracies of the LN detection were 0.93 on the validation dataset (0.83 on the test dataset) compared to 0.81 (0.81) on the validation (test) datasets when using the standard method of thresholding U-Net predictions to arrive at a binary mask. Our method allowed for the creation of an "uncertain" category, and partly limited false-positive predictions on the external dataset. The mean Dice score was 0.73 (0.60) per-image and 0.66 (0.48) per-LN for the validation (test) datasets. Our pipeline detects images with LNs more accurately than conventional methods, and high-throughput delineation of LNs can facilitate future LN content analyses of large datasets.

Investigation of the added value of CT-based radiomics in predicting the development of brain metastases in patients with radically treated stage III NSCLC

Introduction:

Despite radical intent therapy for patients with stage III non-small-cell lung cancer (NSCLC), cumulative incidence of brain metastases (BM) reaches 30%. Current risk stratification methods fail to accurately identify these patients. As radiomics features have been shown to have predictive value, this study aims to develop a model combining clinical risk factors with radiomics features for BM development in patients with radically treated stage III NSCLC.

Methods:

Retrospective analysis of two prospective multicentre studies. Inclusion criteria: adequately staged [¹⁸F-fluorodeoxyglucose positron emission tomography-computed tomography (18-FDG-PET-CT), contrast-enhanced chest CT, contrast-enhanced brain magnetic resonance imaging/CT] and radically treated stage III NSCLC, exclusion criteria: second primary within 2 years of NSCLC diagnosis and prior prophylactic cranial irradiation. Primary endpoint was BM development any time during follow-up (FU). CT-based radiomics features (N = 530) were extracted from the primary lung tumour on 18-FDG-PET-CT images, and a list of clinical features (N = 8) was collected. Univariate feature selection based on the area under the curve (AUC) of the receiver operating characteristic was performed to identify relevant features. Generalized linear models were trained using the selected features, and multivariate predictive performance was assessed through the AUC.

Results:

In total, 219 patients were eligible for analysis. Median FU was 59.4 months for the training cohort and 67.3 months for the validation cohort; 21 (15%) and 17 (22%) patients developed BM in the training and validation cohort, respectively. Two relevant clinical features (age and adenocarcinoma histology) and four relevant radiomics features were identified as predictive. The clinical model yielded the highest AUC value of 0.71 (95% CI: 0.58–0.84), better than radiomics or a combination of clinical parameters and radiomics (both an AUC of 0.62, 95% CIs of 0.47–076 and 0.48–0.76, respectively).

Conclusion:

CT-based radiomics features of primary NSCLC in the current setup could not improve on a model based on clinical predictors (age and adenocarcinoma histology) of BM development in radically treated stage III NSCLC patients.

Predicting Adverse Radiation Effects in Brain Tumors After Stereotactic Radiotherapy With Deep Learning and Handcrafted Radiomics

Introduction

There is a cumulative risk of 20–40% of developing brain metastases (BM) in solid cancers. Stereotactic radiotherapy (SRT) enables the application of high focal doses of radiation to a volume and is often used for BM treatment. However, SRT can cause adverse radiation effects (ARE), such as radiation necrosis, which sometimes cause irreversible damage to the brain. It is therefore of clinical interest to identify patients at a high risk of developing ARE. We hypothesized that models trained with radiomics features, deep learning (DL) features, and patient characteristics or their combination can predict ARE risk in patients with BM before SRT.

Methods

Gadolinium-enhanced T1-weighted MRIs and characteristics from patients treated with SRT for BM were collected for a training and testing cohort (N = 1,404) and a validation cohort (N = 237) from a separate institute. From each lesion in the training set, radiomics features were extracted and used to train an extreme gradient boosting (XGBoost) model. A DL model was trained on the same cohort to make a separate prediction and to extract the last layer of features. Different models using XGBoost were built using only radiomics features, DL features, and patient characteristics or a combination of them. Evaluation was performed using the area under the curve (AUC) of the receiver operating characteristic curve on the external dataset. Predictions for individual lesions and per patient developing ARE were investigated.

Results

The best-performing XGBoost model on a lesion level was trained on a combination of radiomics features and DL features (AUC of 0.71 and recall of 0.80). On a patient level, a combination of radiomics features, DL features, and patient characteristics obtained the best performance (AUC of 0.72 and recall of 0.84). The DL model achieved an AUC of 0.64 and recall of 0.85 per lesion and an AUC of 0.70 and recall of 0.60 per patient.

Conclusion

Machine learning models built on radiomics features and DL features extracted from BM combined with patient characteristics show potential to predict ARE at the patient and lesion levels. These models could be used in clinical decision making, informing patients on their risk of ARE and allowing physicians to opt for different therapies.

Diagnosis of Idiopathic Pulmonary Fibrosis in High-Resolution Computed Tomography Scans Using a Combination of Handcrafted Radiomics and Deep Learning

Purpose

To develop handcrafted radiomics (HCR) and deep learning (DL) based automated diagnostic tools that can differentiate between idiopathic pulmonary fibrosis (IPF) and non-IPF interstitial lung diseases (ILDs) in patients using high-resolution computed tomography (HRCT) scans.

Material and Methods

In this retrospective study, 474 HRCT scans were included (mean age, 64.10 years ± 9.57 [SD]). Five-fold cross-validation was performed on 365 HRCT scans. Furthermore, an external dataset comprising 109 patients was used as a test set. An HCR model, a DL model, and an ensemble of HCR and DL model were developed. A virtual in-silico trial was conducted with two radiologists and one pulmonologist on the same external test set for performance comparison. The performance was compared using DeLong method and McNemar test. Shapley Additive exPlanations (SHAP) plots and Grad-CAM heatmaps were used for the post-hoc interpretability of HCR and DL models, respectively.

Results

In five-fold cross-validation, the HCR model, DL model, and the ensemble of HCR and DL models achieved accuracies of 76.2 ± 6.8, 77.9 ± 4.6, and 85.2 ± 2.7%, respectively. For the diagnosis of IPF and non-IPF ILDs on the external test set, the HCR, DL, and the ensemble of HCR and DL models achieved accuracies of 76.1, 77.9, and 85.3%, respectively. The ensemble model outperformed the diagnostic performance of clinicians who achieved a mean accuracy of 66.3 ± 6.7% (p < 0.05) during the in-silico trial. The area under the receiver operating characteristic curve (AUC) for the ensemble model on the test set was 0.917 which was significantly higher than the HCR model (0.817, p = 0.02) and the DL model (0.823, p = 0.005). The agreement between HCR and DL models was 61.4%, and the accuracy and specificity for the predictions when both the models agree were 93 and 97%, respectively. SHAP analysis showed the texture features as the most important features for IPF diagnosis and Grad-CAM showed that the model focused on the clinically relevant part of the image.

Conclusion

Deep learning and HCR models can complement each other and serve as useful clinical aids for the diagnosis of IPF and non-IPF ILDs.

Automated detection and segmentation of non-small cell lung cancer computed tomography images

Detection and segmentation of abnormalities on medical images is highly important for patient management including diagnosis, radiotherapy, response evaluation, as well as for quantitative image research. We present a fully automated pipeline for the detection and volumetric segmentation of non-small cell lung cancer (NSCLC) developed and validated on 1328 thoracic CT scans from 8 institutions. Along with quantitative performance detailed by image slice thickness, tumor size, image interpretation difficulty, and tumor location, we report an in-silico prospective clinical trial, where we show that the proposed method is faster and more reproducible compared to the experts. Moreover, we demonstrate that on average, radiologists & radiation oncologists preferred automatic segmentations in 56% of the cases. Additionally, we evaluate the prognostic power of the automatic contours by applying RECIST criteria and measuring the tumor volumes. Segmentations by our method stratified patients into low and high survival groups with higher significance compared to those methods based on manual contours.

Correct interpretation of computer tomography (CT) scans is important for the correct assessment of a patient’s disease but can be subjective and timely. Here, the authors develop a system that can automatically segment the non-small cell lung cancer on CT images of patients and show in an in silico trial that the method was faster and more reproducible than clinicians.

Machine learning-based combined nomogram for predicting the risk of pulmonary invasive fungal infection in severely immunocompromised patients

Background

Early and accurate diagnosis of invasive fungal infection (IFI) is pivotal for the initiation of effective antifungal therapy for patients with hematologic malignancies.

Methods

This retrospective study involved 235 patients with hematologic malignancies and pulmonary infections diagnosed as IFIs (n=118) or bacterial pneumonia (n=117). Patients were randomly divided into training (n=188) and validation (n=47) datasets. Four feature selection methods with nine classifiers were implemented to select the optimal machine learning (ML) model using five-fold cross-validation. A radiomic signature was constructed using a linear ML algorithm, and a radiomic score (Radscore) was calculated. The combined model was developed with the Radscore, the significant clinical and radiologic factors were selected using multivariable logistic regression, and the results were presented as a clinical radiomic nomogram. A prospective pilot study was also conducted to compare the classification performance of the combined nomogram with practicing radiologists.

Results

Significant differences were found in the Radscore between IFI and bacterial pneumonia patients in the training (0.683 vs. −0.724, P<0.001) and validation set (0.353 vs. −0.717, P=0.002). The combined model showed good discrimination performance in the validation cohort [area under the curve (AUC) =0.844] and outperformed the clinical (AUC =0.696) and radiomics (AUC =0.767) model alone (both P<0.05).

Conclusions

The clinical radiomic nomogram can serve as a promising predictive tool for IFI in patients with hematologic malignancies.

Comprehensive summary and retrospective evaluation of prognostic scores for patients with newly diagnosed brain metastases treated with upfront radiosurgery in a modern patient collective

BACKGROUND

Numerous prognostic scores (PS) for patients with brain metastases (BM) have been developed. Recently, PS based on laboratory parameters were introduced to better predict overall survival (OS). A comprehensive comparison of the wide range of scores in a modern patient collective is still missing.

MATERIALS AND METHODS

Twelve PS considering clinical parameters only at the time of BM diagnosis were calculated for 470 patients receiving upfront SRS between January 2014 and March 2020. In a subcohort of 310 patients where a full laboratory dataset was available five additional prognostic scores were compared. Restricted mean survival time (RMST), partial likelihood and c-index were calculated as metrics for performance evaluation. Univariable and multivariable analysis were used to identify prognostic factors for OS.

RESULTS

The median OS of the whole cohort was 15.8 months (95% C.I.: 13.4-20.1). All prognostic scores performed well in separating patients into different prognostic groups. RPA achieved the highest c-index, whereas GGS achieved highest partial likelihood with evaluation in the total cohort. With incorporation of the laboratory scores the recently suggested EC-GPA achieved highest c-index and highest partial likelihood. A prognostic score solely based on the assessment of performance status achieved considerable high performance as either 3- or 4-tiered score. Multivariable analysis revealed performance status, systemic disease status and laboratory parameters to be significantly associated with OS among variates included in prognostic scores.

CONCLUSION

Although recent PS incorporating laboratory parameters show convincing performance in predicting overall survival, older scores relying on clinical parameters only are still valid and appealing as they are easier to calculate, and as overall performance is almost equal. Moreover, a score just based on performance status is not significantly inferior and should at least be assessed for informed decision making.

A Handcrafted Radiomics-Based Model for the Diagnosis of Usual Interstitial Pneumonia in Patients with Idiopathic Pulmonary Fibrosis

The most common idiopathic interstitial lung disease (ILD) is idiopathic pulmonary fibrosis (IPF). It can be identified by the presence of usual interstitial pneumonia (UIP) via high-resolution computed tomography (HRCT) or with the use of a lung biopsy. We hypothesized that a CT-based approach using handcrafted radiomics might be able to identify IPF patients with a radiological or histological UIP pattern from those with an ILD or normal lungs. A total of 328 patients from one center and two databases participated in this study. Each participant had their lungs automatically contoured and sectorized. The best radiomic features were selected for the random forest classifier and performance was assessed using the area under the receiver operator characteristics curve (AUC). A significant difference in the volume of the trachea was seen between a normal state, IPF, and non-IPF ILD. Between normal and fibrotic lungs, the AUC of the classification model was 1.0 in validation. When classifying between IPF with a typical HRCT UIP pattern and non-IPF ILD the AUC was 0.96 in validation. When classifying between IPF with UIP (radiological or biopsy-proved) and non-IPF ILD, an AUC of 0.66 was achieved in the testing dataset. Classification between normal, IPF/UIP, and other ILDs using radiomics could help discriminate between different types of ILDs via HRCT, which are hardly recognizable with visual assessments. Radiomic features could become a valuable tool for computer-aided decision-making in imaging, and reduce the need for unnecessary biopsies.

Transparency of deep neural networks for medical image analysis: A review of interpretability methods

Artificial Intelligence (AI) has emerged as a useful aid in numerous clinical applications for diagnosis and treatment decisions. Deep neural networks have shown the same or better performance than clinicians in many tasks owing to the rapid increase in the available data and computational power. In order to conform to the principles of trustworthy AI, it is essential that the AI system be transparent, robust, fair, and ensure accountability. Current deep neural solutions are referred to as black-boxes due to a lack of understanding of the specifics concerning the decision-making process. Therefore, there is a need to ensure the interpretability of deep neural networks before they can be incorporated into the routine clinical workflow. In this narrative review, we utilized systematic keyword searches and domain expertise to identify nine different types of interpretability methods that have been used for understanding deep learning models for medical image analysis applications based on the type of generated explanations and technical similarities. Furthermore, we report the progress made towards evaluating the explanations produced by various interpretability methods. Finally, we discuss limitations, provide guidelines for using interpretability methods and future directions concerning the interpretability of deep neural networks for medical imaging analysis.

Test-Retest Data for the Assessment of Breast MRI Radiomic Feature Repeatability

Background

Radiomic features extracted from breast MRI have potential for diagnostic, prognostic, and predictive purposes. However, before they can be used as biomarkers in clinical decision support systems, features need to be repeatable and reproducible.

Objective

Identify repeatable radiomics features within breast tissue on prospectively collected MRI exams through multiple test–retest measurements.

Study Type

Prospective.

Population

11 healthy female volunteers.

Field Strength/Sequence

1.5 T; MRI exams, comprising T2‐weighted turbo spin‐echo (T2W) sequence, native T1‐weighted turbo gradient‐echo (T1W) sequence, diffusion‐weighted imaging (DWI) sequence using b‐values 0/150/800, and corresponding derived ADC maps.

Assessment

18 MRI exams (three test–retest settings, repeated on 2 days) per healthy volunteer were examined on an identical scanner using a fixed clinical breast protocol. For each scan, 91 features were extracted from the 3D manually segmented right breast using Pyradiomics, before and after image preprocessing. Image preprocessing consisted of 1) bias field correction (BFC); 2) z‐score normalization with and without BFC; 3) grayscale discretization using 32 and 64 bins with and without BFC; and 4) z‐score normalization + grayscale discretization using 32 and 64 bins with and without BFC.

Statistical Tests

Features' repeatability was assessed using concordance correlation coefficient(CCC) for each pair, i.e. each MRI was compared to each of the remaining 17 MRI with a cut‐off value of CCC > 0.90.

Results

Images without preprocessing produced the highest number of repeatable features for both T1W sequence and ADC maps with 15 of 91 (16.5%) and 8 of 91 (8.8%) repeatable features, respectively. Preprocessed images produced between 4 of 91 (4.4%) and 14 of 91 (15.4%), and 6 of 91 (6.6%) and 7 of 91 (7.7%) repeatable features, respectively for T1W and ADC maps. Z‐score normalization produced highest number of repeatable features, 26 of 91 (28.6%) in T2W sequences, in these images, no preprocessing produced 11 of 91 (12.1%) repeatable features.

Data Conclusion

Radiomic features extracted from T1W, T2W sequences and ADC maps from breast MRI exams showed a varying number of repeatable features, depending on the sequence. Effects of different preprocessing procedures on repeatability of features were different for each sequence.

Level of Evidence

2

Technical Efficacy Stage

1

Machine learning for grading and prognosis of esophageal dysplasia using mass spectrometry and histological imaging

BACKGROUND

Barrett's esophagus (BE) is a precursor lesion of esophageal adenocarcinoma and may progress from non-dysplastic through low-grade dysplasia (LGD) to high-grade dysplasia (HGD) and cancer. Grading BE is of crucial prognostic value and is currently based on the subjective evaluation of biopsies. This study aims to investigate the potential of machine learning (ML) using spatially resolved molecular data from mass spectrometry imaging (MSI) and histological data from microscopic hematoxylin and eosin (H&E)-stained imaging for computer-aided diagnosis and prognosis of BE.

METHODS

Biopsies from 57 patients were considered, divided into non-dysplastic (n = 15), LGD non-progressive (n = 14), LGD progressive (n = 14), and HGD (n = 14). MSI experiments were conducted at 50 × 50 μm spatial resolution per pixel corresponding to a tile size of 96x96 pixels in the co-registered H&E images, making a total of 144,823 tiles for the whole dataset.

RESULTS

ML models were trained to distinguish epithelial tissue from stroma with area-under-the-curve (AUC) values of 0.89 (MSI) and 0.95 (H&E)) and dysplastic grade (AUC of 0.97 (MSI) and 0.85 (H&E)) on a tile level, and low-grade progressors from non-progressors on a patient level (accuracies of 0.72 (MSI) and 0.48 (H&E)).

CONCLUSIONS

In summary, while the H&E-based classifier was best at distinguishing tissue types, the MSI-based model was more accurate at distinguishing dysplastic grades and patients at progression risk, which demonstrates the complementarity of both approaches. Data are available via ProteomeXchange with identifier PXD028949.

MRI-based delta-radiomics predicts pathologic complete response in high-grade soft-tissue sarcoma patients treated with neoadjuvant therapy

PURPOSE

In high-grade soft-tissue sarcomas (STS) the standard of care encompasses multimodal therapy regimens. While there is a growing body of evidence for prognostic pretreatment radiomic models, we hypothesized that temporal changes in radiomic features following neoadjuvant treatment ("delta-radiomics") may be able to predict the pathological complete response (pCR).

METHODS

MRI scans (T1-weighted with fat-saturation and contrast-enhancement (T1FSGd) and T2-weighted with fat-saturation (T2FS)) of patients with STS of the extremities and trunk treated with neoadjuvant therapy were gathered from two independent institutions (training: 103, external testing: 53 patients). pCR was defined as <5% viable cells. After segmentation and preprocessing, 105 radiomic features were extracted. Delta-radiomic features were calculated by subtraction of features derived from MRI scans obtained before and after neoadjuvant therapy. After feature reduction, machine learning modeling was performed in 100 iterations of 3-fold nested cross-validation. Delta-radiomic models were compared with single timepoint models in the testing cohort.

RESULTS

The combined delta-radiomic models achieved the best area under the receiver operating characteristic curve (AUC) of 0.75. Pre-therapeutic tumor volume was the best conventional predictor (AUC 0.70). The T2FS-based delta-radiomic model had the most balanced classification performance with a balanced accuracy of 0.69. Delta-radiomic models achieved better reproducibility than single timepoint radiomic models, RECIST or the peri-therapeutic volume change. Delta-radiomic models were significantly associated with survival in multivariate Cox regression.

CONCLUSION

This exploratory analysis demonstrated that MRI-based delta-radiomics improves prediction of pCR over tumor volume and RECIST. Delta-radiomics may one day function as a biomarker for personalized treatment adaptations.

Reproducibility of CT-Based Hepatocellular Carcinoma Radiomic Features across Different Contrast Imaging Phases: A Proof of Concept on SORAMIC Trial Data

Simple Summary

Radiomics has been reported to have potential for correlating with clinical outcomes. However, handcrafted radiomic features (HRFs)—the quantitative features extracted from medical images—are limited by their sensitivity to variations in scanning parameters. Furthermore, radiomics analyses require big data with good quality to achieve desirable performances. In this study, we investigated the reproducibility of HRFs between scans acquired with the same scanning parameters except for the imaging phase (arterial and portal venous phases) to assess the possibilities of merging scans from different phases or replacing missing scans from a phase with other phases to increase data entries. Additionally, we assessed the potential of ComBat harmonization to remove batch effects attributed to this variation. Our results show that the majority of HRFs were not reproducible between the arterial and portal venous phases before or after ComBat harmonization. We provide a guide for analyzing scans of different imaging phases.

Abstract

Handcrafted radiomic features (HRFs) are quantitative imaging features extracted from regions of interest on medical images which can be correlated with clinical outcomes and biologic characteristics. While HRFs have been used to train predictive and prognostic models, their reproducibility has been reported to be affected by variations in scan acquisition and reconstruction parameters, even within the same imaging vendor. In this work, we evaluated the reproducibility of HRFs across the arterial and portal venous phases of contrast-enhanced computed tomography images depicting hepatocellular carcinomas, as well as the potential of ComBat harmonization to correct for this difference. ComBat harmonization is a method based on Bayesian estimates that was developed for gene expression arrays, and has been investigated as a potential method for harmonizing HRFs. Our results show that the majority of HRFs are not reproducible between the arterial and portal venous imaging phases, yet a number of HRFs could be used interchangeably between those phases. Furthermore, ComBat harmonization increased the number of reproducible HRFs across both phases by 1%. Our results guide the pooling of arterial and venous phases from different patients in an effort to increase cohort size, as well as joint analysis of the phases.

Making Radiomics More Reproducible across Scanner and Imaging Protocol Variations: A Review of Harmonization Methods

Radiomics converts medical images into mineable data via a high-throughput extraction of quantitative features used for clinical decision support. However, these radiomic features are susceptible to variation across scanners, acquisition protocols, and reconstruction settings. Various investigations have assessed the reproducibility and validation of radiomic features across these discrepancies. In this narrative review, we combine systematic keyword searches with prior domain knowledge to discuss various harmonization solutions to make the radiomic features more reproducible across various scanners and protocol settings. Different harmonization solutions are discussed and divided into two main categories: image domain and feature domain. The image domain category comprises methods such as the standardization of image acquisition, post-processing of raw sensor-level image data, data augmentation techniques, and style transfer. The feature domain category consists of methods such as the identification of reproducible features and normalization techniques such as statistical normalization, intensity harmonization, ComBat and its derivatives, and normalization using deep learning. We also reflect upon the importance of deep learning solutions for addressing variability across multi-centric radiomic studies especially using generative adversarial networks (GANs), neural style transfer (NST) techniques, or a combination of both. We cover a broader range of methods especially GANs and NST methods in more detail than previous reviews.

Exploratory Radiomic Analysis of Conventional vs. Quantitative Brain MRI: Toward Automatic Diagnosis of Early Multiple Sclerosis

Conventional magnetic resonance imaging (cMRI) is poorly sensitive to pathological changes related to multiple sclerosis (MS) in normal-appearing white matter (NAWM) and gray matter (GM), with the added difficulty of not being very reproducible. Quantitative MRI (qMRI), on the other hand, attempts to represent the physical properties of tissues, making it an ideal candidate for quantitative medical image analysis or radiomics. We therefore hypothesized that qMRI-based radiomic features have added diagnostic value in MS compared to cMRI. This study investigated the ability of cMRI (T1w) and qMRI features extracted from white matter (WM), NAWM, and GM to distinguish between MS patients (MSP) and healthy control subjects (HCS). We developed exploratory radiomic classification models on a dataset comprising 36 MSP and 36 HCS recruited in CHU Liege, Belgium, acquired with cMRI and qMRI. For each image type and region of interest, qMRI radiomic models for MS diagnosis were developed on a training subset and validated on a testing subset. Radiomic models based on cMRI were developed on the entire training dataset and externally validated on open-source datasets with 167 HCS and 10 MSP. Ranked by region of interest, the best diagnostic performance was achieved in the whole WM. Here the model based on magnetization transfer imaging (a type of qMRI) features yielded a median area under the receiver operating characteristic curve (AUC) of 1.00 in the testing sub-cohort. Ranked by image type, the best performance was achieved by the magnetization transfer models, with median AUCs of 0.79 (0.69–0.90, 90% CI) in NAWM and 0.81 (0.71–0.90) in GM. The external validation of the T1w models yielded an AUC of 0.78 (0.47–1.00) in the whole WM, demonstrating a large 95% CI and a low sensitivity of 0.30 (0.10–0.70). This exploratory study indicates that qMRI radiomics could provide efficient diagnostic information using NAWM and GM analysis in MSP. T1w radiomics could be useful for a fast and automated check of conventional MRI for WM abnormalities once acquisition and reconstruction heterogeneities have been overcome. Further prospective validation is needed, involving more data for better interpretation and generalization of the results.

Privacy preserving distributed learning classifiers - Sequential learning with small sets of data

BACKGROUND

Artificial intelligence (AI) typically requires a significant amount of high-quality data to build reliable models, where gathering enough data within a single institution can be particularly challenging. In this study we investigated the impact of using sequential learning to exploit very small, siloed sets of clinical and imaging data to train AI models. Furthermore, we evaluated the capacity of such models to achieve equivalent performance when compared to models trained with the same data over a single centralized database.

METHODS

We propose a privacy preserving distributed learning framework, learning sequentially from each dataset. The framework is applied to three machine learning algorithms: Logistic Regression, Support Vector Machines (SVM), and Perceptron. The models were evaluated using four open-source datasets (Breast cancer, Indian liver, NSCLC-Radiomics dataset, and Stage III NSCLC).

FINDINGS

The proposed framework ensured a comparable predictive performance against a centralized learning approach. Pairwise DeLong tests showed no significant difference between the compared pairs for each dataset.

INTERPRETATION

Distributed learning contributes to preserve medical data privacy. We foresee this technology will increase the number of collaborative opportunities to develop robust AI, becoming the default solution in scenarios where collecting enough data from a single reliable source is logistically impossible. Distributed sequential learning provides privacy persevering means for institutions with small but clinically valuable datasets to collaboratively train predictive AI while preserving the privacy of their patients. Such models perform similarly to models that are built on a larger central dataset.

A Prospectively Validated Prognostic Model for Patients with Locally Advanced Squamous Cell Carcinoma of the Head and Neck Based on Radiomics of Computed Tomography Images

BACKGROUND

Locoregionally advanced head and neck squamous cell carcinoma (HNSCC) patients have high relapse and mortality rates. Imaging-based decision support may improve outcomes by optimising personalised treatment, and support patient risk stratification. We propose a multifactorial prognostic model including radiomics features to improve risk stratification for advanced HNSCC, compared to TNM eighth edition, the gold standard.

PATIENT AND METHODS

Data of 666 retrospective- and 143 prospective-stage III-IVA/B HNSCC patients were collected. A multivariable Cox proportional-hazards model was trained to predict overall survival (OS) using diagnostic CT-based radiomics features extracted from the primary tumour. Separate analyses were performed using TNM8, tumour volume, clinical and biological variables, and combinations thereof with radiomics features. Patient risk stratification in three groups was assessed through Kaplan-Meier (KM) curves. A log-rank test was performed for significance (p-value < 0.05). The prognostic accuracy was reported through the concordance index (CI).

RESULTS

A model combining an 11-feature radiomics signature, clinical and biological variables, TNM8, and volume could significantly stratify the validation cohort into three risk groups (p < 0∙01, CI of 0.79 as validation).

CONCLUSION

A combination of radiomics features with other predictors can predict OS very accurately for advanced HNSCC patients and improves on the current gold standard of TNM8.

Development and External Validation of Deep-Learning-Based Tumor Grading Models in Soft-Tissue Sarcoma Patients Using MR Imaging

BACKGROUND

In patients with soft-tissue sarcomas, tumor grading constitutes a decisive factor to determine the best treatment decision. Tumor grading is obtained by pathological work-up after focal biopsies. Deep learning (DL)-based imaging analysis may pose an alternative way to characterize STS tissue. In this work, we sought to non-invasively differentiate tumor grading into low-grade (G1) and high-grade (G2/G3) STS using DL techniques based on MR-imaging.

METHODS

Contrast-enhanced T1-weighted fat-saturated (T1FSGd) MRI sequences and fat-saturated T2-weighted (T2FS) sequences were collected from two independent retrospective cohorts (training: 148 patients, testing: 158 patients). Tumor grading was determined following the French Federation of Cancer Centers Sarcoma Group in pre-therapeutic biopsies. DL models were developed using transfer learning based on the DenseNet 161 architecture.

RESULTS

The T1FSGd and T2FS-based DL models achieved area under the receiver operator characteristic curve (AUC) values of 0.75 and 0.76 on the test cohort, respectively. T1FSGd achieved the best F1-score of all models (0.90). The T2FS-based DL model was able to significantly risk-stratify for overall survival. Attention maps revealed relevant features within the tumor volume and in border regions.

CONCLUSIONS

MRI-based DL models are capable of predicting tumor grading with good reproducibility in external validation.

Modeling-Based Decision Support System for Radical Prostatectomy Versus External Beam Radiotherapy for Prostate Cancer Incorporating an In Silico Clinical Trial and a Cost-Utility Study

Simple Summary

Low–intermediate prostate cancer has a number of viable treatment options, such as radical prostatectomy and radiotherapy, with similar survival outcomes but different treatment-related side effects. The aim of this study is to facilitate patient-specific treatment selection by developing a decision support system (DSS) that incorporates predictive models for cancer-free survival and treatment-related side effects. We challenged this DSS by validating it against randomized clinical trials and assessing the benefit through a cost–utility analysis. We aim to expand upon the applications of this DSS by using it as the basis for an in silico clinical trial for an underrepresented patient group. This modeling study shows that DSS-based treatment decisions will result in a clinically relevant increase in the patients’ quality of life and can be used for in silico trials.

Abstract

The aim of this study is to build a decision support system (DSS) to select radical prostatectomy (RP) or external beam radiotherapy (EBRT) for low- to intermediate-risk prostate cancer patients. We used an individual state-transition model based on predictive models for estimating tumor control and toxicity probabilities. We performed analyses on a synthetically generated dataset of 1000 patients with realistic clinical parameters, externally validated by comparison to randomized clinical trials, and set up an in silico clinical trial for elderly patients. We assessed the cost-effectiveness (CE) of the DSS for treatment selection by comparing it to randomized treatment allotment. Using the DSS, 47.8% of synthetic patients were selected for RP and 52.2% for EBRT. During validation, differences with the simulations of late toxicity and biochemical failure never exceeded 2%. The in silico trial showed that for elderly patients, toxicity has more influence on the decision than TCP, and the predicted QoL depends on the initial erectile function. The DSS is estimated to result in cost savings (EUR 323 (95% CI: EUR 213–433)) and more quality-adjusted life years (QALYs; 0.11 years, 95% CI: 0.00–0.22) than randomized treatment selection.

MRI-Based Radiomics Analysis for the Pretreatment Prediction of Pathologic Complete Tumor Response to Neoadjuvant Systemic Therapy in Breast Cancer Patients: A Multicenter Study

This retrospective study investigated the value of pretreatment contrast-enhanced Magnetic Resonance Imaging (MRI)-based radiomics for the prediction of pathologic complete tumor response to neoadjuvant systemic therapy in breast cancer patients. A total of 292 breast cancer patients, with 320 tumors, who were treated with neo-adjuvant systemic therapy and underwent a pretreatment MRI exam were enrolled. As the data were collected in two different hospitals with five different MRI scanners and varying acquisition protocols, three different strategies to split training and validation datasets were used. Radiomics, clinical, and combined models were developed using random forest classifiers in each strategy. The analysis of radiomics features had no added value in predicting pathologic complete tumor response to neoadjuvant systemic therapy in breast cancer patients compared with the clinical models, nor did the combined models perform significantly better than the clinical models. Further, the radiomics features selected for the models and their performance differed with and within the different strategies. Due to previous and current work, we tentatively attribute the lack of improvement in clinical models following the addition of radiomics to the effects of variations in acquisition and reconstruction parameters. The lack of reproducibility data (i.e., test-retest or similar) meant that this effect could not be analyzed. These results indicate the need for reproducibility studies to preselect reproducible features in order to properly assess the potential of radiomics.

The application of a workflow integrating the variable reproducibility and harmonizability of radiomic features on a phantom dataset

Radiomics–the high throughput extraction of quantitative features from medical images and their correlation with clinical and biological endpoints- is the subject of active and extensive research. Although the field shows promise, the generalizability of radiomic signatures is affected significantly by differences in scan acquisition and reconstruction settings. Previous studies reported on the sensitivity of radiomic features (RFs) to test-retest variability, inter-observer segmentation variability, and intra-scanner variability. A framework involving robust radiomics analysis and the application of a post-reconstruction feature harmonization method using ComBat was recently proposed to address these challenges. In this study, we investigated the reproducibility of RFs across different scanners and scanning parameters using this framework. We analysed thirteen scans of a ten-layer phantom that were acquired differently. Each layer was subdivided into sixteen regions of interest (ROIs), and the scans were compared in a pairwise manner, resulting in seventy-eight different scenarios. Ninety-one RFs were extracted from each ROI. As hypothesized, we demonstrate that the reproducibility of a given RF is not a constant but is dependent on the heterogeneity found in the data under analysis. The number (%) of reproducible RFs varied across the pairwise scenarios investigated, having a wide range between 8 (8.8%) and 78 (85.7%) RFs. Furthermore, in contrast to what has been previously reported, and as hypothesized in the robust radiomics analysis framework, our results demonstrate that ComBat cannot be applied to all RFs but rather on a percentage of those–the “ComBatable” RFs–which differed depending on the data being harmonized. The number (%) of reproducible RFs following ComBat harmonization varied across the pairwise scenarios investigated, ranging from 14 (15.4%) to 80 (87.9%) RFs, and was found to depend on the heterogeneity in the data. We conclude that the standardization of image acquisition protocols remains the cornerstone for improving the reproducibility of RFs, and the generalizability of the signatures developed. Our proposed approach helps identify the reproducible RFs across different datasets.

Prognostic Assessment in High-Grade Soft-Tissue Sarcoma Patients: A Comparison of Semantic Image Analysis and Radiomics

BACKGROUND

In patients with soft-tissue sarcomas of the extremities, the treatment decision is currently regularly based on tumor grading and size. The imaging-based analysis may pose an alternative way to stratify patients' risk. In this work, we compared the value of MRI-based radiomics with expert-derived semantic imaging features for the prediction of overall survival (OS).

METHODS

Fat-saturated T2-weighted sequences (T2FS) and contrast-enhanced T1-weighted fat-saturated (T1FSGd) sequences were collected from two independent retrospective cohorts (training: 108 patients; testing: 71 patients). After preprocessing, 105 radiomic features were extracted. Semantic imaging features were determined by three independent radiologists. Three machine learning techniques (elastic net regression (ENR), least absolute shrinkage and selection operator, and random survival forest) were compared to predict OS.

RESULTS

ENR models achieved the best predictive performance. Histologies and clinical staging differed significantly between both cohorts. The semantic prognostic model achieved a predictive performance with a C-index of 0.58 within the test set. This was worse compared to a clinical staging system (C-index: 0.61) and the radiomic models (C-indices: T1FSGd: 0.64, T2FS: 0.63). Both radiomic models achieved significant patient stratification.

CONCLUSIONS

T2FS and T1FSGd-based radiomic models outperformed semantic imaging features for prognostic assessment.

The Effects of In-Plane Spatial Resolution on CT-Based Radiomic Features' Stability with and without ComBat Harmonization

Simple Summary

Handcrafted radiomic features (HRFs) are quantitative features extracted from medical images, and they are mined for associations with different clinical endpoints. While many studies reported on the potential of HRFs to unravel clinical endpoints, the sensitivity of HRFs to variations in scanning parameters is affecting the inclusion of radiomic signatures in clinical decision-making. In this study, we investigated the effects of variations in the in-plane resolution of scans, while all other scanning parameters were fixed. Moreover, we investigated the effects of ten different image resampling methods and ComBat harmonization on the reproducibility of HRFs. Our results show that the majority of HRFs are significantly and variably affected by the differences in in-plane resolution. The majority of image resampling methods resulted in a higher number of reproducible HRFs compared to ComBat harmonization. Our developed framework guides identifying the reproducible and harmonizable HRFs in different scenarios.

Abstract

While handcrafted radiomic features (HRFs) have shown promise in the field of personalized medicine, many hurdles hinder its incorporation into clinical practice, including but not limited to their sensitivity to differences in acquisition and reconstruction parameters. In this study, we evaluated the effects of differences in in-plane spatial resolution (IPR) on HRFs, using a phantom dataset (n = 14) acquired on two scanner models. Furthermore, we assessed the effects of interpolation methods (IMs), the choice of a new unified in-plane resolution (NUIR), and ComBat harmonization on the reproducibility of HRFs. The reproducibility of HRFs was significantly affected by variations in IPR, with pairwise concordant HRFs, as measured by the concordance correlation coefficient (CCC), ranging from 42% to 95%. The number of concordant HRFs (CCC > 0.9) after resampling varied depending on (i) the scanner model, (ii) the IM, and (iii) the NUIR. The number of concordant HRFs after ComBat harmonization depended on the variations between the batches harmonized. The majority of IMs resulted in a higher number of concordant HRFs compared to ComBat harmonization, and the combination of IMs and ComBat harmonization did not yield a significant benefit. Our developed framework can be used to assess the reproducibility and harmonizability of RFs.

Development and external validation of a non-invasive molecular status predictor of chromosome 1p/19q co-deletion based on MRI radiomics analysis of Low Grade Glioma patients

PURPOSE

The 1p/19q co-deletion status has been demonstrated to be a prognostic biomarker in lower grade glioma (LGG). The objective of this study was to build a magnetic resonance (MRI)-derived radiomics model to predict the 1p/19q co-deletion status.

METHOD

209 pathology-confirmed LGG patients from 2 different datasets from The Cancer Imaging Archive were retrospectively reviewed; one dataset with 159 patients as the training and discovery dataset and the other one with 50 patients as validation dataset. Radiomics features were extracted from T2- and T1-weighted post-contrast MRI resampled data using linear and cubic interpolation methods. For each of the voxel resampling methods a three-step approach was used for feature selection and a random forest (RF) classifier was trained on the training dataset. Model performance was evaluated on training and validation datasets and clinical utility indexes (CUIs) were computed. The distributions and intercorrelation for selected features were analyzed.

RESULTS

Seven radiomics features were selected from the cubic interpolated features and five from the linear interpolated features on the training dataset. The RF classifier showed similar performance for cubic and linear interpolation methods in the training dataset with accuracies of 0.81 (0.75-0.86) and 0.76 (0.71-0.82) respectively; in the validation dataset the accuracy dropped to 0.72 (0.6-0.82) using cubic interpolation and 0.72 (0.6-0.84) using linear resampling. CUIs showed the model achieved satisfactory negative values (0.605 using cubic interpolation and 0.569 for linear interpolation).

CONCLUSIONS

MRI has the potential for predicting the 1p/19q status in LGGs. Both cubic and linear interpolation methods showed similar performance in external validation.

Structural and functional radiomics for lung cancer

INTRODUCTION

Lung cancer ranks second in new cancer cases and first in cancer-related deaths worldwide. Precision medicine is working on altering treatment approaches and improving outcomes in this patient population. Radiological images are a powerful non-invasive tool in the screening and diagnosis of early-stage lung cancer, treatment strategy support, prognosis assessment, and follow-up for advanced-stage lung cancer. Recently, radiological features have evolved from solely semantic to include (handcrafted and deep) radiomic features. Radiomics entails the extraction and analysis of quantitative features from medical images using mathematical and machine learning methods to explore possible ties with biology and clinical outcomes.

METHODS

Here, we outline the latest applications of both structural and functional radiomics in detection, diagnosis, and prediction of pathology, gene mutation, treatment strategy, follow-up, treatment response evaluation, and prognosis in the field of lung cancer.

CONCLUSION

The major drawbacks of radiomics are the lack of large datasets with high-quality data, standardization of methodology, the black-box nature of deep learning, and reproducibility. The prerequisite for the clinical implementation of radiomics is that these limitations are addressed. Future directions include a safer and more efficient model-training mode, merge multi-modality images, and combined multi-discipline or multi-omics to form "Medomics."

Cycle-Consistent Generative Adversarial Network: Effect on Radiation Dose Reduction and Image Quality Improvement in Ultralow-Dose CT for Evaluation of Pulmonary Tuberculosis

Objective

To investigate the image quality of ultralow-dose CT (ULDCT) of the chest reconstructed using a cycle-consistent generative adversarial network (CycleGAN)-based deep learning method in the evaluation of pulmonary tuberculosis.

Materials and Methods

Between June 2019 and November 2019, 103 patients (mean age, 40.8 ± 13.6 years; 61 men and 42 women) with pulmonary tuberculosis were prospectively enrolled to undergo standard-dose CT (120 kVp with automated exposure control), followed immediately by ULDCT (80 kVp and 10 mAs). The images of the two successive scans were used to train the CycleGAN framework for image-to-image translation. The denoising efficacy of the CycleGAN algorithm was compared with that of hybrid and model-based iterative reconstruction. Repeated-measures analysis of variance and Wilcoxon signed-rank test were performed to compare the objective measurements and the subjective image quality scores, respectively.

Results

With the optimized CycleGAN denoising model, using the ULDCT images as input, the peak signal-to-noise ratio and structural similarity index improved by 2.0 dB and 0.21, respectively. The CycleGAN-generated denoised ULDCT images typically provided satisfactory image quality for optimal visibility of anatomic structures and pathological findings, with a lower level of image noise (mean ± standard deviation [SD], 19.5 ± 3.0 Hounsfield unit [HU]) than that of the hybrid (66.3 ± 10.5 HU, p < 0.001) and a similar noise level to model-based iterative reconstruction (19.6 ± 2.6 HU, p > 0.908). The CycleGAN-generated images showed the highest contrast-to-noise ratios for the pulmonary lesions, followed by the model-based and hybrid iterative reconstruction. The mean effective radiation dose of ULDCT was 0.12 mSv with a mean 93.9% reduction compared to standard-dose CT.

Conclusion

The optimized CycleGAN technique may allow the synthesis of diagnostically acceptable images from ULDCT of the chest for the evaluation of pulmonary tuberculosis.

Dedicated Axillary MRI-Based Radiomics Analysis for the Prediction of Axillary Lymph Node Metastasis in Breast Cancer

Radiomics features may contribute to increased diagnostic performance of MRI in the prediction of axillary lymph node metastasis. The objective of the study was to predict preoperative axillary lymph node metastasis in breast cancer using clinical models and radiomics models based on T2-weighted (T2W) dedicated axillary MRI features with node-by-node analysis. From August 2012 until October 2014, all women who had undergone dedicated axillary 3.0T T2W MRI, followed by axillary surgery, were retrospectively identified, and available clinical data were collected. All axillary lymph nodes were manually delineated on the T2W MR images, and quantitative radiomics features were extracted from the delineated regions. Data were partitioned patient-wise to train 100 models using different splits for the training and validation cohorts to account for multiple lymph nodes per patient and class imbalance. Features were selected in the training cohorts using recursive feature elimination with repeated 5-fold cross-validation, followed by the development of random forest models. The performance of the models was assessed using the area under the curve (AUC). A total of 75 women (median age, 61 years; interquartile range, 51-68 years) with 511 axillary lymph nodes were included. On final pathology, 36 (7%) of the lymph nodes had metastasis. A total of 105 original radiomics features were extracted from the T2W MR images. Each cohort split resulted in a different number of lymph nodes in the training cohorts and a different set of selected features. Performance of the 100 clinical and radiomics models showed a wide range of AUC values between 0.41-0.74 and 0.48-0.89 in the training cohorts, respectively, and between 0.30-0.98 and 0.37-0.99 in the validation cohorts, respectively. With these results, it was not possible to obtain a final prediction model. Clinical characteristics and dedicated axillary MRI-based radiomics with node-by-node analysis did not contribute to the prediction of axillary lymph node metastasis in breast cancer based on data where variations in acquisition and reconstruction parameters were not addressed.

Non-invasive imaging prediction of tumor hypoxia: A novel developed and externally validated CT and FDG-PET-based radiomic signatures

BACKGROUND

Tumor hypoxia increases resistance to radiotherapy and systemic therapy. Our aim was to develop and validate a disease-agnostic and disease-specific CT (+FDG-PET) based radiomics hypoxia classification signature.

MATERIAL AND METHODS

A total of 808 patients with imaging data were included: N = 100 training/N = 183 external validation cases for a disease-agnostic CT hypoxia classification signature, N = 76 training/N = 39 validation cases for the H&N CT signature and N = 62 training/N = 36 validation cases for the Lung CT signature. The primary gross tumor volumes (GTV) were manually defined by experts on CT. In order to dichotomize between hypoxic/well-oxygenated tumors a threshold of 20% was used for the [¹⁸F]-HX4-derived hypoxic fractions (HF). A random forest (RF)-based machine-learning classifier/regressor was trained to classify patients as hypoxia-positive/ negative based on radiomic features.

RESULTS

A 11 feature "disease-agnostic CT model" reached AUC's of respectively 0.78 (95% confidence interval [CI], 0.62-0.94), 0.82 (95% CI, 0.67-0.96) and 0.78 (95% CI, 0.67-0.89) in three external validation datasets. A "disease-agnostic FDG-PET model" reached an AUC of 0.73 (0.95% CI, 0.49-0.97) in validation by combining 5 features. The highest "lung-specific CT model" reached an AUC of 0.80 (0.95% CI, 0.65-0.95) in validation with 4 CT features, while the "H&N-specific CT model" reached an AUC of 0.84 (0.95% CI, 0.64-1.00) in validation with 15 CT features. A tumor volume-alone model was unable to significantly classify patients as hypoxia-positive/ negative. A significant survival split (P = 0.037) was found between CT-classified hypoxia strata in an external H&N cohort (n = 517), while 117 significant hypoxia gene-CT signature feature associations were found in an external lung cohort (n = 80).

CONCLUSION

The disease-specific radiomics signatures perform better than the disease agnostic ones. By identifying hypoxic patients our signatures have the potential to enrich interventional hypoxia-targeting trials.

Implementing Systems Modelling and Molecular Imaging to Predict the Efficacy of BCL-2 Inhibition in Colorectal Cancer Patient-Derived Xenograft Models

Resistance to chemotherapy often results from dysfunctional apoptosis, however multiple proteins with overlapping functions regulate this pathway. We sought to determine whether an extensively validated, deterministic apoptosis systems model, 'DR_MOMP', could be used as a stratification tool for the apoptosis sensitiser and BCL-2 antagonist, ABT-199 in patient-derived xenograft (PDX) models of colorectal cancer (CRC). Through quantitative profiling of BCL-2 family proteins, we identified two PDX models which were predicted by DR_MOMP to be sufficiently sensitive to 5-fluorouracil (5-FU)-based chemotherapy (CRC0344), or less responsive to chemotherapy but sensitised by ABT-199 (CRC0076). Treatment with ABT-199 significantly improved responses of CRC0076 PDXs to 5-FU-based chemotherapy, but showed no sensitisation in CRC0344 PDXs, as predicted from systems modelling. ¹⁸F-Fluorodeoxyglucose positron emission tomography/computed tomography (¹⁸F-FDG-PET/CT) scans were performed to investigate possible early biomarkers of response. In CRC0076, a significant post-treatment decrease in mean standard uptake value was indeed evident only in the combination treatment group. Radiomic CT feature analysis of pre-treatment images in CRC0076 and CRC0344 PDXs identified features which could phenotypically discriminate between models, but were not predictive of treatment responses. Collectively our data indicate that systems modelling may identify metastatic (m)CRC patients benefitting from ABT-199, and that ¹⁸F-FDG-PET could independently support such predictions.

Diagnosis of Invasive Lung Adenocarcinoma Based on Chest CT Radiomic Features of Part-Solid Pulmonary Nodules: A Multicenter Study

Background Solid components of part-solid nodules (PSNs) at CT are reflective of invasive adenocarcinoma, but studies describing radiomic features of PSNs and the perinodular region are lacking. Purpose To develop and to validate radiomic signatures diagnosing invasive lung adenocarcinoma in PSNs compared with the Brock, clinical-semantic features, and volumetric models. Materials and Methods This retrospective multicenter study (https://ClinicalTrials.gov, NCT03872362) included 291 patients (median age, 60 years; interquartile range, 55-65 years; 191 women) from January 2013 to October 2017 with 297 PSN lung adenocarcinomas split into training (n = 229) and test (n = 68) data sets. Radiomic features were extracted from the different regions (gross tumor volume [GTV], solid, ground-glass, and perinodular). Random-forest models were trained using clinical-semantic, volumetric, and radiomic features, and an online nodule calculator was used to compute the Brock model. Performances of models were evaluated using standard metrics such as area under the curve (AUC), accuracy, and calibration. The integrated discrimination improvement was applied to assess model performance changes after the addition of perinodular features. Results The radiomics model based on ground-glass and solid features yielded an AUC of 0.98 (95% confidence interval [CI]: 0.96, 1.00) on the test data set, which was significantly higher than the Brock (AUC, 0.83 [95% CI: 0.72, 0.94]; P = .007), clinical-semantic (AUC, 0.90 [95% CI: 0.83, 0.98]; P = .03), volumetric GTV (AUC, 0.87 [95% CI: 0.78, 0.96]; P = .008), and radiomics GTV (AUC, 0.88 [95% CI: 0.80, 0.96]; P = .01) models. It also achieved the best accuracy (93% [95% CI: 84%, 98%]). Both this model and the model with added perinodular features showed good calibration, whereas adding perinodular features did not improve the performance (integrated discrimination improvement, -0.02; P = .56). Conclusion Separating ground-glass and solid CT radiomic features of part-solid nodules was useful in diagnosing the invasiveness of lung adenocarcinoma, yielding a better predictive performance than the Brock, clinical-semantic, volumetric, and radiomics gross tumor volume models. Online supplemental material is available for this article. See also the editorial by Nishino in this issue. Published under a CC BY 4.0 license.

Development of a clinical decision support system for severity risk prediction and triage of COVID-19 patients at hospital admission: an international multicentre study

BACKGROUND

The outbreak of coronavirus disease 2019 (COVID-19) has globally strained medical resources and caused significant mortality.

OBJECTIVE

To develop and validate a machine-learning model based on clinical features for severity risk assessment and triage for COVID-19 patients at hospital admission.

METHOD

725 patients were used to train and validate the model. This included a retrospective cohort from Wuhan, China of 299 hospitalised COVID-19 patients from 23 December 2019 to 13 February 2020, and five cohorts with 426 patients from eight centres in China, Italy and Belgium from 20 February 2020 to 21 March 2020. The main outcome was the onset of severe or critical illness during hospitalisation. Model performances were quantified using the area under the receiver operating characteristic curve (AUC) and metrics derived from the confusion matrix.

RESULTS

In the retrospective cohort, the median age was 50 years and 137 (45.8%) were male. In the five test cohorts, the median age was 62 years and 236 (55.4%) were male. The model was prospectively validated on five cohorts yielding AUCs ranging from 0.84 to 0.93, with accuracies ranging from 74.4% to 87.5%, sensitivities ranging from 75.0% to 96.9%, and specificities ranging from 55.0% to 88.0%, most of which performed better than the pneumonia severity index. The cut-off values of the low-, medium- and high-risk probabilities were 0.21 and 0.80. The online calculators can be found at www.covid19risk.ai.

CONCLUSION

The machine-learning model, nomogram and online calculator might be useful to access the onset of severe and critical illness among COVID-19 patients and triage at hospital admission.

Correction: Computed tomography-derived radiomic signature of head and neck squamous cell carcinoma (peri)tumoral tissue for the prediction of locoregional recurrence and distant metastasis after concurrent chemo-radiotherapy

[This corrects the article DOI: 10.1371/journal.pone.0232639.].