Artificial intelligence: Deep learning in oncological radiomics and challenges of interpretability and data harmonization

Comparing quantitative image parameters between animal and clinical CT-scanners: a translational phantom study analysis

Purpose

This study compares phantom-based variability of extracted radiomics features from scans on a photon counting CT (PCCT) and an experimental animal PET/CT-scanner (Albira II) to investigate the potential of radiomics for translation from animal models to human scans. While oncological basic research in animal PET/CT has allowed an intrinsic comparison between PET and CT, but no 1:1 translation to a human CT scanner due to resolution and noise limitations, Radiomics as a statistical and thus scale-independent method can potentially close the critical gap.

Methods

Two phantoms were scanned on a PCCT and animal PET/CT-scanner with different scan parameters and then the radiomics parameters were extracted. A Principal Component Analysis (PCA) was conducted. To overcome the limitation of a small dataset, a data augmentation technique was applied. A Ridge Classifier was trained and a Feature Importance- and Cluster analysis was performed.

Results

PCA and Cluster Analysis shows a clear differentiation between phantom types while emphasizing the comparability of both scanners. The Ridge Classifier exhibited a strong training performance with 93% accuracy, but faced challenges in generalization with a test accuracy of 62%.

Conclusion

These results show that radiomics has great potential as a translational tool between animal models and human routine diagnostics, especially using the novel photon counting technique. This is another crucial step towards integration of radiomics analysis into clinical practice.

The radiomorphological appearance of the invasive margin in pancreatic cancer is associated with tumor budding

PURPOSE

Pancreatic cancer (PDAC) is characterized by infiltrative, spiculated tumor growth into the surrounding non-neoplastic tissue. Clinically, its diagnosis is often established by magnetic resonance imaging (MRI). At the invasive margin, tumor buds can be detected by histology, an established marker associated with poor prognosis in different types of tumors.

METHODS

We analyzed PDAC by determining the degree of tumor spiculation on T2-weighted MRI using a 3-tier grading system. The grade of spiculation was correlated with the density of tumor buds quantified in histological sections of the respective surgical specimen according to the guidelines of the International Tumor Budding Consensus Conference (n = 28 patients).

RESULTS

64% of tumors revealed intermediate to high spiculation on MRI. In over 90% of cases, tumor buds were detected. We observed a significant positive rank correlation between the grade of radiological tumor spiculation and the histopathological number of tumor buds (rs = 0.745, p < 0.001). The number of tumor buds was not significantly associated with tumor stage, presence of lymph node metastases, or histopathological grading (p ≥ 0.352).

CONCLUSION

Our study identifies a readily available radiological marker for non-invasive estimation of tumor budding, as a correlate for infiltrative tumor growth. This finding could help to identify PDAC patients who might benefit from more extensive peripancreatic soft tissue resection during surgery or stratify patients for personalized therapy concepts.

Distinguishing between aldosterone-producing adenomas and non-functional adrenocortical adenomas using the YOLOv5 network

BACKGROUND

You Only Look Once version 5 (YOLOv5), a one-stage deep-learning (DL) algorithm for object detection and classification, offers high speed and accuracy for identifying targets.

PURPOSE

To investigate the feasibility of using the YOLOv5 algorithm to non-invasively distinguish between aldosterone-producing adenomas (APAs) and non-functional adrenocortical adenomas (NF-ACAs) on computed tomography (CT) images.

MATERIAL AND METHODS

A total of 235 patients who were diagnosed with ACAs between January 2011 and July 2022 were included in this study. Of the 215 patients, 81 (37.7%) had APAs and 134 (62.3%) had NF-ACAs' they were randomly divided into either the training set or the validation set at a ratio of 9:1. Another 20 patients, including 8 (40.0%) with APA and 12 (60.0%) with NF-ACA, were collected for the testing set. Five submodels (YOLOv5n, YOLOv5s, YOLOv5m, YOLOv5l, and YOLOv5x) of YOLOv5 were trained and evaluated on the datasets.

RESULTS

In the testing set, the mAP_0.5 value for YOLOv5x (0.988) was higher than the values for YOLOv5n (0.969), YOLOv5s (0.965), YOLOv5m (0.974), and YOLOv5l (0.983). The mAP_0.5:0.95 value for YOLOv5x (0.711) was also higher than the values for YOLOv5n (0.587), YOLOv5s (0.674), YOLOv5m (0.671), and YOLOv5l (0.698) in the testing set. The inference speed of YOLOv5n was 2.4 ms in the testing set, which was the fastest among the five submodels.

CONCLUSION

The YOLOv5 algorithm can accurately and efficiently distinguish between APAs and NF-ACAs on CT images, especially YOLOv5x has the best identification performance.

Investigation of a potential upstream harmonization based on image appearance matching to improve radiomics features robustness: a phantom study

Objective. Radiomics is a promising valuable analysis tool consisting in extracting quantitative information from medical images. However, the extracted radiomics features are too sensitive to variations in used image acquisition and reconstruction parameters. This limited robustness hinders the generalizable validity of radiomics-assisted models. Our aim is to investigate a possible harmonization strategy based on matching image quality to improve feature robustness.Approach.We acquired CT scans of a phantom with two scanners across different dose levels and percentages of Iterative Reconstruction algorithms. The detectability index was used as a comprehensive task-based image quality metric. A statistical analysis based on the Intraclass Correlation Coefficient was performed to determine if matching image quality/appearance could enhance the robustness of radiomics features extracted from the phantom images. Additionally, an Artificial Neural Network was trained on these features to automatically classify the scanner used for image acquisition.Main results.We found that the ICC of the features across protocols providing a similar detectability index improves with respect to the ICC of the features across protocols providing a different detectability index. This improvement was particularly noticeable in features relevant for distinguishing between scanners.Significance.This preliminary study demonstrates that a harmonization based on image quality/appearance matching could improve radiomics features robustness and heterogeneous protocols can be used to obtain a similar image appearance in terms of the detectability index. Thus protocols with a lower dose level could be selected to reduce the amount of radiation dose delivered to the patient and simultaneously obtain a more robust quantitative analysis.

BraNet: a mobil application for breast image classification based on deep learning algorithms

Mobile health apps are widely used for breast cancer detection using artificial intelligence algorithms, providing radiologists with second opinions and reducing false diagnoses. This study aims to develop an open-source mobile app named "BraNet" for 2D breast imaging segmentation and classification using deep learning algorithms. During the phase off-line, an SNGAN model was previously trained for synthetic image generation, and subsequently, these images were used to pre-trained SAM and ResNet18 segmentation and classification models. During phase online, the BraNet app was developed using the react native framework, offering a modular deep-learning pipeline for mammography (DM) and ultrasound (US) breast imaging classification. This application operates on a client-server architecture and was implemented in Python for iOS and Android devices. Then, two diagnostic radiologists were given a reading test of 290 total original RoI images to assign the perceived breast tissue type. The reader's agreement was assessed using the kappa coefficient. The BraNet App Mobil exhibited the highest accuracy in benign and malignant US images (94.7%/93.6%) classification compared to DM during training I (80.9%/76.9%) and training II (73.7/72.3%). The information contrasts with radiological experts' accuracy, with DM classification being 29%, concerning US 70% for both readers, because they achieved a higher accuracy in US ROI classification than DM images. The kappa value indicates a fair agreement (0.3) for DM images and moderate agreement (0.4) for US images in both readers. It means that not only the amount of data is essential in training deep learning algorithms. Also, it is vital to consider the variety of abnormalities, especially in the mammography data, where several BI-RADS categories are present (microcalcifications, nodules, mass, asymmetry, and dense breasts) and can affect the API accuracy model.

Revolutionizing tumor detection and classification in multimodality imaging based on deep learning approaches: methods, applications and limitations

BACKGROUND

The emergence of deep learning (DL) techniques has revolutionized tumor detection and classification in medical imaging, with multimodal medical imaging (MMI) gaining recognition for its precision in diagnosis, treatment, and progression tracking.

OBJECTIVE

This review comprehensively examines DL methods in transforming tumor detection and classification across MMI modalities, aiming to provide insights into advancements, limitations, and key challenges for further progress.

METHODS

Systematic literature analysis identifies DL studies for tumor detection and classification, outlining methodologies including convolutional neural networks (CNNs), recurrent neural networks (RNNs), and their variants. Integration of multimodality imaging enhances accuracy and robustness.

RESULTS

Recent advancements in DL-based MMI evaluation methods are surveyed, focusing on tumor detection and classification tasks. Various DL approaches, including CNNs, YOLO, Siamese Networks, Fusion-Based Models, Attention-Based Models, and Generative Adversarial Networks, are discussed with emphasis on PET-MRI, PET-CT, and SPECT-CT.

FUTURE DIRECTIONS

The review outlines emerging trends and future directions in DL-based tumor analysis, aiming to guide researchers and clinicians toward more effective diagnosis and prognosis. Continued innovation and collaboration are stressed in this rapidly evolving domain.

CONCLUSION

Conclusions drawn from literature analysis underscore the efficacy of DL approaches in tumor detection and classification, highlighting their potential to address challenges in MMI analysis and their implications for clinical practice.

A novel approach to identify the brain regions that best classify ADHD by means of EEG and deep learning

Objective

Attention-Deficit Hyperactivity Disorder (ADHD) is one of the most widespread neurodevelopmental disorders diagnosed in childhood. ADHD is diagnosed by following the guidelines of Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5). According to DSM-5, ADHD has not yet identified a specific cause, and thus researchers continue to investigate this field. Therefore, the primary objective of this work is to present a study to find the subset of channels or brain regions that best classify ADHD vs Typically Developing children by means of Electroencephalograms (EEG).

Methods

To achieve this goal, we present a novel approach to identify the brain regions that best classify ADHD using EEG and Deep Learning (DL). First, we perform a filtering and artefact removal process on the EEG signal. Then we generate different subsets of EEG channels depending on their location on the scalp (hemispheres, lobes, sets of lobes and single channels) and using backward and forward stepwise feature selection methods. Finally, we feed the DL neural network with each set, and compute the f 1 -score.

Results and conclusions

Based on the obtained results, the Frontal Lobe (FL) (0.8081 f 1 -score) and the Left Hemisphere (LH) (0.8056 f 1 -score) provide more significant information detecting individuals with ADHD, than using the entire set of EEG Channels (0.8067 f 1 -score). However, when combining the Temporal, Parietal and Occipital Lobes (TL, PL, OL), better results (0.8097 f 1 -score) were obtained compared with using only the FL and LH subsets. The best performance was obtained using Feature Selection Methods. In the case of the Backward Stepwise Feature Selection method, a combination of 14 EEG channels yielded a 0.8281 f 1 -score. Similarly, using the Forward Stepwise Feature Selection method, a combination of 11 EEG channels yielded a 0.8271 f 1 -score. These findings hold significant value for physicians in the quest to better understand the underlying causes of ADHD.

Markerless vision-based functional movement screening movements evaluation with deep neural networks

The functional movement screen (FMS) test is a seven-test battery used to assess fundamental movement abilities of individuals. It is commonly used to predict sports injuries but relies on clinical expertise and is not suitable for self-examination. This study presents an automatic FMS movement assessment framework using a multi-view deep neural network called MVDNN. The framework combines automatic skeleton extraction with manual feature selection to extract 3D trajectory features of human skeleton joints from two different directions. Three mainstream methods of time-series modeling are then used to learn high-level feature representation from skeleton sequences, and motion features from two views are fused to provide complementary information. Results of public FMS movements dataset demonstrate that our MVDNN outperforms current state-of-the-art methods with an average miF1 score of 0.857, maF1 score of 0.768, and Kappa score of 0.640 over ten runs.

Enhancing Interoperability and Harmonisation of Nuclear Medicine Image Data and Associated Clinical Data

Nuclear imaging techniques such as positron emission tomography (PET) and single photon emission computed tomography (SPECT) in combination with computed tomography (CT) are established imaging modalities in clinical practice, particularly for oncological problems. Due to a multitude of manufacturers, different measurement protocols, local demographic or clinical workflow variations as well as various available reconstruction and analysis software, very heterogeneous datasets are generated. This review article examines the current state of interoperability and harmonisation of image data and related clinical data in the field of nuclear medicine. Various approaches and standards to improve data compatibility and integration are discussed. These include, for example, structured clinical history, standardisation of image acquisition and reconstruction as well as standardised preparation of image data for evaluation. Approaches to improve data acquisition, storage and analysis will be presented. Furthermore, approaches are presented to prepare the datasets in such a way that they become usable for projects applying artificial intelligence (AI) (machine learning, deep learning, etc.). This review article concludes with an outlook on future developments and trends related to AI in nuclear medicine, including a brief research of commercial solutions.

Combined radiomics of primary tumour and bone metastasis improve the prediction of EGFR mutation status and response to EGFR-TKI therapy for NSCLC

PURPOSE

To develop radiomics models of primary tumour and spinal metastases to predict epidermal growth factor receptor (EGFR) mutations and therapeutic response to EGFR-tyrosine kinase inhibitor (TKI) in patients with metastatic non-small-cell lung cancer (NSCLC).

METHODS

We enrolled 203 patients with spinal metastases between December 2017 and September 2021, classified as patients with the EGFR mutation or EGFR wild-type. All patients underwent thoracic CT and spinal MRI scans before any treatment. Radiomics analysis was performed to extract features from primary tumour and metastases images and identify predictive features with the least absolute shrinkage and selection operator. Radiomics signatures (RS) were constructed based on primary tumour (RS-Pri), metastases (RS-Met), and in combination (RS-Com) to predict EGFR mutation status and response to EGFR-TKI. Receiver operating characteristic (ROC) curve analysis with 10-fold cross-validation was applied to assess the performance of the models.

RESULTS

To predict the EGFR mutation status, the RS based on the combination of primary tumour and metastases improved the prediction AUCs compared to those based on the primary tumour or metastasis alone in the training (RS-Com-EGFR: 0.927) and validation (RS-Com-EGFR: 0.812) cohorts. To predict response to EGFR-TKI, the developed RS based on combined primary tumour and metastasis generated the highest AUCs in the training (RS-Com-TKI: 0.880) and validation (RS-Com-TKI: 0.798) cohort.

CONCLUSIONS

Primary NSCLC and spinal metastases can provide complementary information to predict the EGFR mutation status and response to EGFR-TKI. The developed models that integrate primary lesions and metastases may be potential imaging markers to guide individual treatment decisions.

Enhancing histopathological image classification of invasive ductal carcinoma using hybrid harmonization techniques

This study aims to develop a robust pipeline for classifying invasive ductal carcinomas and benign tumors in histopathological images, addressing variability within and between centers. We specifically tackle the challenge of detecting atypical data and variability between common clusters within the same database. Our feature engineering-based pipeline comprises a feature extraction step, followed by multiple harmonization techniques to rectify intra- and inter-center batch effects resulting from image acquisition variability and diverse patient clinical characteristics. These harmonization steps facilitate the construction of more robust and efficient models. We assess the proposed pipeline's performance on two public breast cancer databases, BreaKHIS and IDCDB, utilizing recall, precision, and accuracy metrics. Our pipeline outperforms recent models, achieving 90-95% accuracy in classifying benign and malignant tumors. We demonstrate the advantage of harmonization for classifying patches from different databases. Our top model scored 94.7% for IDCDB and 95.2% for BreaKHis, surpassing existing feature engineering-based models (92.1% for IDCDB and 87.7% for BreaKHIS) and attaining comparable performance to deep learning models. The proposed feature-engineering-based pipeline effectively classifies malignant and benign tumors while addressing variability within and between centers through the incorporation of various harmonization techniques. Our findings reveal that harmonizing variabilities between patches from different batches directly impacts the learning and testing performance of classification models. This pipeline has the potential to enhance breast cancer diagnosis and treatment and may be applicable to other diseases.

Investigating the value of radiomics stemming from DSC quantitative biomarkers in IDH mutation prediction in gliomas

Objective

This study aims to assess the value of biomarker based radiomics to predict IDH mutation in gliomas. The patient cohort consists of 160 patients histopathologicaly proven of primary glioma (WHO grades 2-4) from 3 different centers.

Methods

To quantify the DSC perfusion signal two different mathematical modeling methods were used (Gamma fitting, leakage correction algorithms) considering the assumptions about the compartments contributing in the blood flow between the extra- and intra vascular space.

Results

The Mean slope of increase (MSI) and the K1 parameter of the bidirectional exchange model exhibited the highest performance with (ACC 74.3% AUROC 74.2%) and (ACC 75% AUROC 70.5%) respectively.

Conclusion

The proposed framework on DSC-MRI radiogenomics in gliomas has the potential of becoming a reliable diagnostic support tool exploiting the mathematical modeling of the DSC signal to characterize IDH mutation status through a more reproducible and standardized signal analysis scheme for facilitating clinical translation.

Artificial Intelligence Applications in Glioma With 1p/19q Co-Deletion: A Systematic Review

As an important genomic marker for oligodendrogliomas, early determination of 1p/19q co-deletion status is critical for guiding therapy and predicting prognosis in patients with glioma. The purpose of this study is to systematically review the literature concerning the magnetic resonance imaging (MRI) with artificial intelligence (AI) methods for predicting 1p/19q co-deletion status in glioma. PubMed, Scopus, Embase, and IEEE Xplore were searched in accordance with the Preferred Reporting Items for systematic reviews and meta-analyses guidelines. Methodological quality of studies was assessed according to the Quality Assessment of Diagnostic Accuracy Studies-2. Finally, 28 studies were included in the quantitative analysis. Diagnostic test accuracy reached an area under the ROC curve of 0.71-0.98 were reported in 24 studies. The remaining four studies with no available AUC provided an accuracy of 0.75-0. 89. The included studies varied widely in terms of imaging sequences, input features, and modeling methods. The current review highlighted that integrating MRI with AI technology is a potential tool for determination 1p/19q status pre-operatively and noninvasively, which can possibly help clinical decision-making. However, the reliability and feasibility of this approach still need to be further validated and improved in a real clinical setting. EVIDENCE LEVEL: 2. TECHNICAL EFFICACY: 2.

A radiomic- and dosiomic-based machine learning regression model for pretreatment planning in 177 Lu-DOTATATE therapy

BACKGROUND

Standardized patient-specific pretreatment dosimetry planning is mandatory in the modern era of nuclear molecular radiotherapy, which may eventually lead to improvements in the final therapeutic outcome. Only a comprehensive definition of a dosage therapeutic window encompassing the range of absorbed doses, that is, helpful without being detrimental can lead to therapy individualization and improved outcomes. As a result, setting absorbed dose safety limits for organs at risk (OARs) requires knowledge of the absorbed dose-effect relationship. Data sets of consistent and reliable inter-center dosimetry findings are required to characterize this relationship.

PURPOSE

We developed and standardized a new pretreatment planning model consisting of a predictive dosimetry procedure for OARs in patients with neuroendocrine tumors (NETs) treated with 177 Lu-DOTATATE (Lutathera). In the retrospective study described herein, we used machine learning (ML) regression algorithms to predict absorbed doses in OARs by exploiting a combination of radiomic and dosiomic features extracted from patients' imaging data.

METHODS

Pretreatment and posttreatment data for 20 patients with NETs treated with 177 Lu-DOTATATE were collected from two clinical centers. A total of 3412 radiomic and dosiomic features were extracted from the patients' computed tomography (CT) scans and dose maps, respectively. All dose maps were generated using Monte Carlo simulations. An ML regression model was designed based on ML algorithms for predicting the absorbed dose in every OAR (liver, left kidney, right kidney, and spleen) before and after the therapy and between each therapy session, thus predicting any possible radiotoxic effects.

RESULTS

We evaluated nine ML regression algorithms. Our predictive model achieved a mean absolute dose error (MAE, in Gy) of 0.61 for the liver, 1.58 for the spleen, 1.30 for the left kidney, and 1.35 for the right kidney between pretherapy 68 Ga-DOTATOC positron emission tomography (PET)/CT and posttherapy 177 Lu-DOTATATE single photon emission (SPECT)/CT scans. Τhe best predictive performance observed was based on the gradient boost for the liver, the left kidney and the right kidney, and on the extra tree regressor for the spleen. Evaluation of the model's performance according to its ability to predict the absorbed dose in each OAR in every possible combination of pretherapy 68 Ga-DOTATOC PET/CT and any posttherapy 177 Lu-DOTATATE treatment cycle SPECT/CT scans as well as any 177 Lu-DOTATATE SPECT/CT treatment cycle and the consequent 177 Lu-DOTATATE SPECT/CT treatment cycle revealed mean absorbed dose differences ranges from -0.55 to 0.68 Gy. Incorporating radiodosiomics features from the 68 Ga-DOTATOC PET/CT and first 177 Lu-DOTATATE SPECT/CT treatment cycle scans further improved the precision and minimized the standard deviation of the predictions in nine out of 12 instances. An average improvement of 57.34% was observed (range: 17.53%-96.12%). However, it's important to note that in three instances (i.e., Ga,C.1 → C3 in spleen and left kidney, and Ga,C.1 → C2 in right kidney) we did not observe an improvement (absolute differences of 0.17, 0.08, and 0.05 Gy, respectively). Wavelet-based features proved to have high correlated predictive value, whereas non-linear-based ML regression algorithms proved to be more capable than the linear-based of producing precise prediction in our case.

CONCLUSIONS

The combination of radiomics and dosiomics has potential utility for personalized molecular radiotherapy (PMR) response evaluation and OAR dose prediction. These radiodosiomic features can potentially provide information on any possible disease recurrence and may be highly useful in clinical decision-making, especially regarding dose escalation issues.

Interpretable radiomics method for predicting human papillomavirus status in oropharyngeal cancer using Bayesian networks

OBJECTIVES

To develop a simple interpretable Bayesian Network (BN) to classify HPV status in patients with oropharyngeal cancer.

METHODS

Two hundred forty-six patients, 216 of whom were HPV positive, were used in this study. We extracted 851 radiomics markers from patients' contrast-enhanced Computed Tomography (CT) images. Mens eX Machina (MXM) approach selected two most relevant predictors: sphericity and max2DDiameterRow. The area under the curve (AUC) demonstrated BN model performance in 30% of the data reserved for testing. A Support Vector Machine (SVM) based method was also implemented for comparison purposes.

RESULTS

The Mens eX Machina (MXM) approach selected two most relevant predictors: sphericity and max2DDiameterRow. Areas under the Curves (AUC) were found 0.78 and 0.72 on the training and test data, respectively. When using support vector machine (SVM) and 25 features, the AUC was found 0.83 on the test data.

CONCLUSIONS

The straightforward structure and power of interpretability of our BN model will help clinicians make treatment decisions and enable the non-invasive detection of HPV status from contrast-enhanced CT images. Higher accuracy can be obtained using more complex structures at the expense of lower interpretability.

ADVANCES IN KNOWLEDGE

Radiomics is being studied lately as a simple imaging data based HPV status detection technique which can be an alternative to laboratory approaches. However, it generally lacks interpretability. This work demonstrated the feasibility of using Bayesian networks based radiomics for predicting HPV positivity in an interpretable way.

Interpretable artificial intelligence in radiology and radiation oncology

Artificial intelligence has been introduced to clinical practice, especially radiology and radiation oncology, from image segmentation, diagnosis, treatment planning and prognosis. It is not only crucial to have an accurate artificial intelligence model, but also to understand the internal logic and gain the trust of the experts. This review is intended to provide some insights into core concepts of the interpretability, the state-of-the-art methods for understanding the machine learning models, the evaluation of these methods, identifying some challenges and limits of them, and gives some examples of medical applications.

Automated segmentation of meningioma from contrast-enhanced T1-weighted MRI images in a case series using a marker-controlled watershed segmentation and fuzzy C-means clustering machine learning algorithm

INTRODUCTION AND IMPORTANCE

Accurate segmentation of meningiomas from contrast-enhanced T1-weighted (CE T1-w) magnetic resonance imaging (MRI) is crucial for diagnosis and treatment planning. Manual segmentation is time-consuming and prone to variability. To evaluate an automated segmentation approach for meningiomas using marker-controlled watershed segmentation (MCWS) and fuzzy c-means (FCM) algorithms.

CASE PRESENTATION AND METHODS

CE T1-w MRI of 3 female patients (aged 59, 44, 67 years) with right frontal meningiomas were analyzed. Images were converted to grayscale and preprocessed with Otsu's thresholding and FCM clustering. MCWS segmentation was performed. Segmentation accuracy was assessed by comparing automated segmentations to manual delineations.

CLINICAL DISCUSSION

The approach successfully segmented meningiomas in all cases. Mean sensitivity was 0.8822, indicating accurate identification of tumors. Mean Dice similarity coefficient between Otsu's and FCM1 was 0.6599, suggesting good overlap between segmentation methods.

CONCLUSION

The MCWS and FCM approach enables accurate automated segmentation of meningiomas from CE T1-w MRI. With further validation on larger datasets, this could provide an efficient tool to assist in delineating meningioma boundaries for clinical management.

A novel breast cancer detection architecture based on a CNN-CBR system for mammogram classification

This paper presents a novel framework for breast cancer detection using mammogram images. The proposed solution aims to output an explainable classification from a mammogram image. The classification approach uses a Case-Based Reasoning system (CBR). CBR accuracy strongly depends on the quality of the extracted features. To achieve relevant classification, we propose a pipeline that includes image enhancement and data augmentation to improve the quality of extracted features and provide a final diagnosis. An efficient segmentation method based on a U-Net architecture is used to extract Regions of interest (RoI) from mammograms. The purpose is to combine deep learning (DL) with CBR to improve classification accuracy. DL provides accurate mammogram segmentation, while CBR gives an explainable and accurate classification. The proposed approach was tested on the CBIS-DDSM dataset and achieved high performance with an accuracy (Acc) of 86.71 % and a recall of 91.34 %, outperforming some well-known machine learning (ML) and DL approaches.

Integrating Artificial Intelligence Tools in the Clinical Research Setting: The Ovarian Cancer Use Case

Artificial intelligence (AI) methods applied to healthcare problems have shown enormous potential to alleviate the burden of health services worldwide and to improve the accuracy and reproducibility of predictions. In particular, developments in computer vision are creating a paradigm shift in the analysis of radiological images, where AI tools are already capable of automatically detecting and precisely delineating tumours. However, such tools are generally developed in technical departments that continue to be siloed from where the real benefit would be achieved with their usage. Significant effort still needs to be made to make these advancements available, first in academic clinical research and ultimately in the clinical setting. In this paper, we demonstrate a prototype pipeline based entirely on open-source software and free of cost to bridge this gap, simplifying the integration of tools and models developed within the AI community into the clinical research setting, ensuring an accessible platform with visualisation applications that allow end-users such as radiologists to view and interact with the outcome of these AI tools.

Morphometry-based radiomics for predicting therapeutic response in patients with gliomas following radiotherapy

Introduction

Gliomas are still considered as challenging in oncologic management despite the developments in treatment approaches. The complete elimination of a glioma might not be possible even after a treatment and assessment of therapeutic response is important to determine the future course of actions for patients with such cancers. In the recent years radiomics has emerged as a promising solution with potential applications including prediction of therapeutic response. Hence, this study was focused on investigating whether morphometry-based radiomics signature could be used to predict therapeutic response in patients with gliomas following radiotherapy.

Methods

105 magnetic resonance (MR) images including segmented and non-segmented images were used to extract morphometric features and develop a morphometry-based radiomics signature. After determining the appropriate machine learning algorithm, a prediction model was developed to predict the therapeutic response eliminating the highly correlated features as well as without eliminating the highly correlated features. Then the model performance was evaluated.

Results

Tumor grade had the highest contribution to develop the morphometry-based signature. Random forest provided the highest accuracy to train the prediction model derived from the morphometry-based radiomics signature. An accuracy of 86% and area under the curve (AUC) value of 0.91 were achieved for the prediction model evaluated without eliminating the highly correlated features whereas accuracy and AUC value were 84% and 0.92 respectively for the prediction model evaluated after eliminating the highly correlated features.

Discussion

Nonetheless, the developed morphometry-based radiomics signature could be utilized as a noninvasive biomarker for therapeutic response in patients with gliomas following radiotherapy.

PET Radiomics and Response to Immunotherapy in Lung Cancer: A Systematic Review of the Literature

The aim of this review is to provide a comprehensive overview of the existing literature concerning the applications of positron emission tomography (PET) radiomics in lung cancer patient candidates or those undergoing immunotherapy.

MATERIALS AND METHODS

A systematic review was conducted on databases and web sources. English-language original articles were considered. The title and abstract were independently reviewed to evaluate study inclusion. Duplicate, out-of-topic, and review papers, or editorials, articles, and letters to editors were excluded. For each study, the radiomics analysis was assessed based on the radiomics quality score (RQS 2.0). The review was registered on the PROSPERO database with the number CRD42023402302.

RESULTS

Fifteen papers were included, thirteen were qualified as using conventional radiomics approaches, and two used deep learning radiomics. The content of each study was different; indeed, seven papers investigated the potential ability of radiomics to predict PD-L1 expression and tumor microenvironment before starting immunotherapy. Moreover, two evaluated the prediction of response, and four investigated the utility of radiomics to predict the response to immunotherapy. Finally, two papers investigated the prediction of adverse events due to immunotherapy.

CONCLUSIONS

Radiomics is promising for the evaluation of TME and for the prediction of response to immunotherapy, but some limitations should be overcome.

Unsupervised Machine Learning of MRI Radiomics Features Identifies Two Distinct Subgroups with Different Liver Function Reserve and Risks of Post-Hepatectomy Liver Failure in Patients with Hepatocellular Carcinoma

OBJECTIVE

To identify subgroups of patients with hepatocellular carcinoma (HCC) with different liver function reserves using an unsupervised machine-learning approach on the radiomics features from preoperative gadoxetic-acid-enhanced MRIs and to evaluate their association with the risk of post-hepatectomy liver failure (PHLF).

METHODS

Clinical data from 276 consecutive HCC patients who underwent liver resections between January 2017 and March 2019 were retrospectively collected. Radiomics features were extracted from the non-tumorous liver tissue at the gadoxetic-acid-enhanced hepatobiliary phase MRI. The reproducible and non-redundant features were selected for consensus clustering analysis to detect distinct subgroups. After that, clinical variables were compared between the identified subgroups to evaluate the clustering efficacy. The liver function reserve of the subgroups was compared and the correlations between the subgroups and PHLF, postoperative complications, and length of hospital stay were evaluated.

RESULTS

A total of 107 radiomics features were extracted and 37 were selected for unsupervised clustering analysis, which identified two distinct subgroups (138 patients in each subgroup). Compared with subgroup 1, subgroup 2 had significantly more patients with older age, albumin-bilirubin grades 2 and 3, a higher indocyanine green retention rate, and a lower indocyanine green plasma disappearance rate (all p < 0.05). Subgroup 2 was also associated with a higher risk of PHLF, postoperative complications, and longer hospital stays (>18 days) than that of subgroup 1, with an odds ratio of 2.83 (95% CI: 1.58-5.23), 2.41(95% CI: 1.15-5.35), and 2.14 (95% CI: 1.32-3.47), respectively. The odds ratio of our method was similar to the albumin-bilirubin grade for postoperative complications and length of hospital stay (2.41 vs. 2.29 and 2.14 vs. 2.16, respectively), but was inferior for PHLF (2.83 vs. 4.55).

CONCLUSIONS

Based on the radiomics features of gadoxetic-acid-enhanced MRI, unsupervised clustering analysis identified two distinct subgroups with different liver function reserves and risks of PHLF in HCC patients. Future studies are required to validate our findings.

Assessment of brain cancer atlas maps with multimodal imaging features

BACKGROUND

Glioblastoma Multiforme (GBM) is a fast-growing and highly aggressive brain tumor that invades the nearby brain tissue and presents secondary nodular lesions across the whole brain but generally does not spread to distant organs. Without treatment, GBM can result in death in about 6 months. The challenges are known to depend on multiple factors: brain localization, resistance to conventional therapy, disrupted tumor blood supply inhibiting effective drug delivery, complications from peritumoral edema, intracranial hypertension, seizures, and neurotoxicity.

MAIN TEXT

Imaging techniques are routinely used to obtain accurate detections of lesions that localize brain tumors. Especially magnetic resonance imaging (MRI) delivers multimodal images both before and after the administration of contrast, which results in displaying enhancement and describing physiological features as hemodynamic processes. This review considers one possible extension of the use of radiomics in GBM studies, one that recalibrates the analysis of targeted segmentations to the whole organ scale. After identifying critical areas of research, the focus is on illustrating the potential utility of an integrated approach with multimodal imaging, radiomic data processing and brain atlases as the main components. The templates associated with the outcome of straightforward analyses represent promising inference tools able to spatio-temporally inform on the GBM evolution while being generalizable also to other cancers.

CONCLUSIONS

The focus on novel inference strategies applicable to complex cancer systems and based on building radiomic models from multimodal imaging data can be well supported by machine learning and other computational tools potentially able to translate suitably processed information into more accurate patient stratifications and evaluations of treatment efficacy.

An interpretable feature-learned model for overall survival classification of High-Grade Gliomas

PURPOSE

An accurate and well-defined survival prediction of High Grade Gliomas (HGGs) is indispensable because of its high incidence and aggressiveness. Therefore, this paper presents a unified framework for fully automatic overall survival classification and its interpretation.

METHODS AND MATERIALS

Initially, a glioma detection model is utilized to detect the tumorous images. A pre-processing module is designed for extracting 2D slices and creating a survival data array for the classification network. Then, the classification pipeline is integrated with two separate pathways: a modality-specific and a modality-concatenated pathway. The modality-specific pathway runs three separate CNNs for extracting rich predictive features from three sub-regions of HGGs (peritumoral edema, enhancing tumor and necrosis) by using three neuro-imaging modalities. In these pathways, the image vectors of the different modalities are also concatenated to the final fusion layer to overcome the loss of lower-level tumor features. Furthermore, to exploit the intra-modality correlations, a modality-concatenated pathway is also added to the classification pipeline. The experiments are conducted on BraTS 2018 and BraTS 2019 benchmarks, demonstrating that the proposed approach performs competitively in classifying HGG patients into three survival groups, namely, short, mid, and long survivors.

RESULTS

The proposed approach achieves an overall classification accuracy, sensitivity, and specificity of about 0.998, 0.997, and 0.999, respectively, for the BraTS 2018 dataset, and for BraTS 2019, these values correspond to 1.000, 0.999, and 0.999.

CONCLUSIONS

The results indicate that the proposed model achieves the highest values of the evaluation metrics for the overall survival classification of HGG.

External validation of an MR-based radiomic model predictive of locoregional control in oropharyngeal cancer

OBJECTIVES

To externally validate a pre-treatment MR-based radiomics model predictive of locoregional control in oropharyngeal squamous cell carcinoma (OPSCC) and to assess the impact of differences between datasets on the predictive performance.

METHODS

Radiomic features, as defined in our previously published radiomics model, were extracted from the primary tumor volumes of 157 OPSCC patients in a different institute. The developed radiomics model was validated using this cohort. Additionally, parameters influencing performance, such as patient subgroups, MRI acquisition, and post-processing steps on prediction performance will be investigated. For this analysis, matched subgroups (based on human papillomavirus (HPV) status of the tumor, T-stage, and tumor subsite) and a subgroup with only patients with 4-mm slice thickness were studied. Also the influence of harmonization techniques (ComBat harmonization, quantile normalization) and the impact of feature stability across observers and centers were studied. Model performances were assessed by area under the curve (AUC), sensitivity, and specificity.

RESULTS

Performance of the published model (AUC/sensitivity/specificity: 0.74/0.75/0.60) drops when applied on the validation cohort (AUC/sensitivity/specificity: 0.64/0.68/0.60). The performance of the full validation cohort improves slightly when the model is validated using a patient group with comparable HPV status of the tumor (AUC/sensitivity/specificity: 0.68/0.74/0.60), using patients acquired with a slice thickness of 4 mm (AUC/sensitivity/specificity: 0.67/0.73/0.57), or when quantile harmonization was performed (AUC/sensitivity/specificity: 0.66/0.69/0.60).

CONCLUSION

The previously published model shows its generalizability and can be applied on data acquired from different vendors and protocols. Harmonization techniques and subgroup definition influence performance of predictive radiomics models.

KEY POINTS

• Radiomics, a noninvasive quantitative image analysis technique, can support the radiologist by enhancing diagnostic accuracy and/or treatment decision-making. • A previously published model shows its generalizability and could be applied on data acquired from different vendors and protocols.

Survey of explainable artificial intelligence techniques for biomedical imaging with deep neural networks

Artificial Intelligence (AI) techniques of deep learning have revolutionized the disease diagnosis with their outstanding image classification performance. In spite of the outstanding results, the widespread adoption of these techniques in clinical practice is still taking place at a moderate pace. One of the major hindrance is that a trained Deep Neural Networks (DNN) model provides a prediction, but questions about why and how that prediction was made remain unanswered. This linkage is of utmost importance for the regulated healthcare domain to increase the trust in the automated diagnosis system by the practitioners, patients and other stakeholders. The application of deep learning for medical imaging has to be interpreted with caution due to the health and safety concerns similar to blame attribution in the case of an accident involving autonomous cars. The consequences of both a false positive and false negative cases are far reaching for patients' welfare and cannot be ignored. This is exacerbated by the fact that the state-of-the-art deep learning algorithms comprise of complex interconnected structures, millions of parameters, and a 'black box' nature, offering little understanding of their inner working unlike the traditional machine learning algorithms. Explainable AI (XAI) techniques help to understand model predictions which help develop trust in the system, accelerate the disease diagnosis, and meet adherence to regulatory requirements. This survey provides a comprehensive review of the promising field of XAI for biomedical imaging diagnostics. We also provide a categorization of the XAI techniques, discuss the open challenges, and provide future directions for XAI which would be of interest to clinicians, regulators and model developers.

Radiomics approach with deep learning for predicting T4 obstructive colorectal cancer using CT image

OBJECTIVES

Patients with T4 obstructive colorectal cancer (OCC) have a high mortality rate. Therefore, an accurate distinction between T4 and T1-T3 (NT4) in OCC is an important part of preoperative evaluation, especially in the emergency setting. This paper introduces three models of radiomics, deep learning, and deep learning-based radiomics to identify T4 OCC.

METHODS

We established a dataset of computed tomography (CT) images of 164 patients with pathologically confirmed OCC, from which 2537 slides were extracted. First, since T4 tumors penetrate the bowel wall and involve adjacent organs, we explored whether the peritumoral region contributes to the assessment of T4 OCC. Furthermore, we visualized the radiomics and deep learning features using the t-distributed stochastic neighbor embedding technique (t-SNE). Finally, we built a merged model by fusing radiomic features with deep learning features. In this experiment, the performance of each model was evaluated by the area under the receiver operating characteristic curve (AUC).

RESULTS

In the test cohort, the AUC values predicted by the radiomics model in the dilated region of interest (dROI) was 0.770. And the AUC value of the deep learning model with the patches extended 20-pixel reached 0.936. Combining the characteristics of radiomics and deep learning, our method achieved an AUC value of 0.947 in the T4 and non-T4 (NT4) classification, and increased the AUC value to 0.950 after the addition of clinical features.

CONCLUSION

The prediction results of our merged model of deep learning radiomics outperformed the deep learning model and significantly outperformed the radiomics model. The experimental results demonstrate that combining the peritumoral region improves the prediction performance of the radiomics model and the deep learning model.

Deriving quantitative information from multiparametric MRI via Radiomics: Evaluation of the robustness and predictive value of radiomic features in the discrimination of low-grade versus high-grade gliomas with machine learning

PURPOSE

Analysis pipelines based on the computation of radiomic features on medical images are widely used exploration tools across a large variety of image modalities. This study aims to define a robust processing pipeline based on Radiomics and Machine Learning (ML) to analyze multiparametric Magnetic Resonance Imaging (MRI) data to discriminate between high-grade (HGG) and low-grade (LGG) gliomas.

METHODS

The dataset consists of 158 multiparametric MRI of patients with brain tumor publicly available on The Cancer Imaging Archive, preprocessed by the BraTS organization committee. Three different types of image intensity normalization algorithms were applied and 107 features were extracted for each tumor region, setting the intensity values according to different discretization levels. The predictive power of radiomic features in the LGG versus HGG categorization was evaluated by using random forest classifiers. The impact of the normalization techniques and of the different settings in the image discretization was studied in terms of the classification performances. A set of MRI-reliable features was defined selecting the features extracted according to the most appropriate normalization and discretization settings.

RESULTS

The results show that using MRI-reliable features improves the performance in glioma grade classification (AUC=0.93±0.05) with respect to the use of raw (AUC=0.88±0.08) and robust features (AUC=0.83±0.08), defined as those not depending on image normalization and intensity discretization.

CONCLUSIONS

These results confirm that image normalization and intensity discretization strongly impact the performance of ML classifiers based on radiomic features. Thus, special attention should be provided in the image preprocessing step before typical radiomic and ML analysis are carried out.

An Automated Method for Artifical Intelligence Assisted Diagnosis of Active Aortitis Using Radiomic Analysis of FDG PET-CT Images

The aim of this study was to develop and validate an automated pipeline that could assist the diagnosis of active aortitis using radiomic imaging biomarkers derived from [18F]-Fluorodeoxyglucose Positron Emission Tomography-Computed Tomography (FDG PET-CT) images. The aorta was automatically segmented by convolutional neural network (CNN) on FDG PET-CT of aortitis and control patients. The FDG PET-CT dataset was split into training (43 aortitis:21 control), test (12 aortitis:5 control) and validation (24 aortitis:14 control) cohorts. Radiomic features (RF), including SUV metrics, were extracted from the segmented data and harmonized. Three radiomic fingerprints were constructed: A-RFs with high diagnostic utility removing highly correlated RFs; B used principal component analysis (PCA); C-Random Forest intrinsic feature selection. The diagnostic utility was evaluated with accuracy and area under the receiver operating characteristic curve (AUC). Several RFs and Fingerprints had high AUC values (AUC > 0.8), confirmed by balanced accuracy, across training, test and external validation datasets. Good diagnostic performance achieved across several multi-centre datasets suggests that a radiomic pipeline can be generalizable. These findings could be used to build an automated clinical decision tool to facilitate objective and standardized assessment regardless of observer experience.

Diffusion radiomics for subtyping and clustering in autism spectrum disorder: A preclinical study

Autism spectrum disorder (ASD) is a highly prevalent, heterogenous neurodevelopmental disorder. Neuroimaging methods such as functional, structural, and diffusion MRI have been used to identify candidate imaging biomarkers for ASD, but current findings remain non-specific and likely arise from the heterogeneity present in ASD. To account for this, efforts to subtype ASD have emerged as a potential strategy for both the study of ASD and advancement of tailored behavioral therapies and therapeutics. Towards these ends, to improve upon current neuroimaging methods, we propose combining biologically sensitive neurite orientation dispersion and density index (NODDI) diffusion MR imaging with radiomics image processing to create a new methodological approach that, we hypothesize, can sensitively and specifically capture neurobiology. We demonstrate this method can sensitively distinguish differences between four genetically distinct rat models of ASD (Fmr1, Pten, Nrxn1, Disc1). Further, we demonstrate diffusion radiomic analyses hold promise for subtyping in ASD as we show unsupervised clustering of NODDI radiomic data generates clusters specific to the underlying genetic differences between the animal models. Taken together, our findings suggest the unique application of radiomic analysis on NODDI diffusion MRI may have the capacity to sensitively and specifically disambiguate the neurobiological heterogeneity present in the ASD population.

Radiomics prognostic analysis of PET/CT images in a multicenter head and neck cancer cohort: investigating ComBat strategies, sub-volume characterization, and automatic segmentation

PURPOSE

This study aimed to investigate the impact of several ComBat harmonization strategies, intra-tumoral sub-volume characterization, and automatic segmentations for progression-free survival (PFS) prediction through radiomics modeling for patients with head and neck cancer (HNC) in PET/CT images.

METHODS

The HECKTOR MICCAI 2021 challenge set containing PET/CT images and clinical data of 325 oropharynx HNC patients was exploited. A total of 346 IBSI-compliant radiomic features were extracted for each patient's primary tumor volume defined by the reference manual contours. Modeling relied on least absolute shrinkage Cox regression (Lasso-Cox) for feature selection (FS) and Cox proportional-hazards (CoxPH) models were built to predict PFS. Within this methodological framework, 8 different strategies for ComBat harmonization were compared, including before or after FS, in feature groups separately or all features directly, and with center or clustering-determined labels. Features extracted from tumor sub-volume clustering were also investigated for their prognostic additional value. Finally, 3 automatic segmentations (2 threshold-based and a 3D U-Net) were also compared. All results were evaluated with the concordance index (C-index).

RESULTS

Radiomics features without harmonization, combined with clinical factors, led to models with C-index values of 0.69 in the testing set. The best version of ComBat harmonization, i.e., after FS, for feature groups separately and relying on clustering-determined labels, achieved a C-index of 0.71. The use of features extracted from tumor sub-volumes further improved the C-index to 0.72. Models that relied on the automatic segmentations yielded close but slightly lower prognostic performance (0.67-0.70) compared to reference contours.

CONCLUSION

A standard radiomics pipeline allowed for prediction of PFS in a multicenter HNC cohort. Applying a specific strategy of ComBat harmonization improved the performance. The extraction of intra-tumoral sub-volume features and automatic segmentation could contribute to the improvement and automation of prognosis modeling, respectively.

Predicting clinically significant prostate cancer with a deep learning approach: a multicentre retrospective study

PURPOSE

This study aimed to develop deep learning (DL) models based on multicentre biparametric magnetic resonance imaging (bpMRI) for the diagnosis of clinically significant prostate cancer (csPCa) and compare the performance of these models with that of the Prostate Imaging and Reporting and Data System (PI-RADS) assessment by expert radiologists based on multiparametric MRI (mpMRI).

METHODS

We included 1861 consecutive male patients who underwent radical prostatectomy or biopsy at seven hospitals with mpMRI. These patients were divided into the training (1216 patients in three hospitals) and external validation cohorts (645 patients in four hospitals). PI-RADS assessment was performed by expert radiologists. We developed DL models for the classification between benign and malignant lesions (DL-BM) and that between csPCa and non-csPCa (DL-CS). An integrated model combining PI-RADS and the DL-CS model, abbreviated as PIDL-CS, was developed. The performances of the DL models and PIDL-CS were compared with that of PI-RADS.

RESULTS

In each external validation cohort, the area under the receiver operating characteristic curve (AUC) values of the DL-BM and DL-CS models were not significantly different from that of PI-RADS (P > 0.05), whereas the AUC of PIDL-CS was superior to that of PI-RADS (P < 0.05), except for one external validation cohort (P > 0.05). The specificity of PIDL-CS for the detection of csPCa was much higher than that of PI-RADS (P < 0.05).

CONCLUSION

Our proposed DL models can be a potential non-invasive auxiliary tool for predicting csPCa. Furthermore, PIDL-CS greatly increased the specificity of csPCa detection compared with PI-RADS assessment by expert radiologists, greatly reducing unnecessary biopsies and helping radiologists achieve a precise diagnosis of csPCa.

Ten quick tips for computational analysis of medical images

Medical imaging is a great asset for modern medicine, since it allows physicians to spatially interrogate a disease site, resulting in precise intervention for diagnosis and treatment, and to observe particular aspect of patients' conditions that otherwise would not be noticeable. Computational analysis of medical images, moreover, can allow the discovery of disease patterns and correlations among cohorts of patients with the same disease, thus suggesting common causes or providing useful information for better therapies and cures. Machine learning and deep learning applied to medical images, in particular, have produced new, unprecedented results that can pave the way to advanced frontiers of medical discoveries. While computational analysis of medical images has become easier, however, the possibility to make mistakes or generate inflated or misleading results has become easier, too, hindering reproducibility and deployment. In this article, we provide ten quick tips to perform computational analysis of medical images avoiding common mistakes and pitfalls that we noticed in multiple studies in the past. We believe our ten guidelines, if taken into practice, can help the computational-medical imaging community to perform better scientific research that eventually can have a positive impact on the lives of patients worldwide.

Utility of adding Radiomics to clinical features in predicting the outcomes of radiotherapy for head and neck cancer using machine learning

BACKGROUND

Radiomics involves the extraction of quantitative information from annotated Computed-Tomography (CT) images, and has been used to predict outcomes in Head and Neck Squamous Cell Carcinoma (HNSCC). Subjecting combined Radiomics and Clinical features to Machine Learning (ML) could offer better predictions of clinical outcomes. This study is a comparative performance analysis of ML models with Clinical, Radiomics, and Clinico-Radiomic datasets for predicting four outcomes of HNSCC treated with Curative Radiation Therapy (RT): Distant Metastases, Locoregional Recurrence, New Primary, and Residual Disease.

METHODOLOGY

The study used retrospective data of 311 HNSCC patients treated with radiotherapy between 2013-2018 at our centre. Binary prediction models were developed for the four outcomes with Clinical-only, Clinico-Radiomic, and Radiomics-only datasets, using three different ML classification algorithms namely, Random Forest (RF), Kernel Support Vector Machine (KSVM), and XGBoost. The best-performing ML algorithms of the three dataset groups was then compared.

RESULTS

The Clinico-Radiomic dataset using KSVM classifier provided the best prediction. Predicted mean testing accuracy for Distant Metastases, Locoregional Recurrence, New Primary, and Residual Disease was 97%, 72%, 99%, and 96%, respectively. The mean area under the receiver operating curve (AUC) was calculated and displayed for all the models using three dataset groups.

CONCLUSION

Clinico-Radiomic dataset improved the predictive ability of ML models over clinical features alone, while models built using Radiomics performed poorly. Radiomics data could therefore effectively supplement clinical data in predicting outcomes.

Application of nnU-Net for Automatic Segmentation of Lung Lesions on CT Images and Its Implication for Radiomic Models

Radiomics investigates the predictive role of quantitative parameters calculated from radiological images. In oncology, tumour segmentation constitutes a crucial step of the radiomic workflow. Manual segmentation is time-consuming and prone to inter-observer variability. In this study, a state-of-the-art deep-learning network for automatic segmentation (nnU-Net) was applied to computed tomography images of lung tumour patients, and its impact on the performance of survival radiomic models was assessed. In total, 899 patients were included, from two proprietary and one public datasets. Different network architectures (2D, 3D) were trained and tested on different combinations of the datasets. Automatic segmentations were compared to reference manual segmentations performed by physicians using the DICE similarity coefficient. Subsequently, the accuracy of radiomic models for survival classification based on either manual or automatic segmentations were compared, considering both hand-crafted and deep-learning features. The best agreement between automatic and manual contours (DICE = 0.78 ± 0.12) was achieved averaging 2D and 3D predictions and applying customised post-processing. The accuracy of the survival classifier (ranging between 0.65 and 0.78) was not statistically different when using manual versus automatic contours, both with hand-crafted and deep features. These results support the promising role nnU-Net can play in automatic segmentation, accelerating the radiomic workflow without impairing the models' accuracy. Further investigations on different clinical endpoints and populations are encouraged to confirm and generalise these findings.

Predictive performance of radiomic models based on features extracted from pretrained deep networks

OBJECTIVES

In radiomics, generic texture and morphological features are often used for modeling. Recently, features extracted from pretrained deep networks have been used as an alternative. However, extracting deep features involves several decisions, and it is unclear how these affect the resulting models. Therefore, in this study, we considered the influence of such choices on the predictive performance.

METHODS

On ten publicly available radiomic datasets, models were trained using feature sets that differed in terms of the utilized network architecture, the layer of feature extraction, the used set of slices, the use of segmentation, and the aggregation method. The influence of these choices on the predictive performance was measured using a linear mixed model. In addition, models with generic features were trained and compared in terms of predictive performance and correlation.

RESULTS

No single choice consistently led to the best-performing models. In the mixed model, the choice of architecture (AUC + 0.016; p < 0.001), the level of feature extraction (AUC + 0.016; p < 0.001), and using all slices (AUC + 0.023; p < 0.001) were highly significant; using the segmentation had a lower influence (AUC + 0.011; p = 0.023), while the aggregation method was insignificant (p = 0.774). Models based on deep features were not significantly better than those based on generic features (p > 0.05 on all datasets). Deep feature sets correlated moderately with each other (r = 0.4), in contrast to generic feature sets (r = 0.89).

CONCLUSIONS

Different choices have a significant effect on the predictive performance of the resulting models; however, for the highest performance, these choices should be optimized during cross-validation.

Artificial intelligence performance in image-based ovarian cancer identification: A systematic review and meta-analysis

BACKGROUND

Accurate identification of ovarian cancer (OC) is of paramount importance in clinical treatment success. Artificial intelligence (AI) is a potentially reliable assistant for the medical imaging recognition. We systematically review articles on the diagnostic performance of AI in OC from medical imaging for the first time.

METHODS

The Medline, Embase, IEEE, PubMed, Web of Science, and the Cochrane library databases were searched for related studies published until August 1, 2022. Inclusion criteria were studies that developed or used AI algorithms in the diagnosis of OC from medical images. The binary diagnostic accuracy data were extracted to derive the outcomes of interest: sensitivity (SE), specificity (SP), and Area Under the Curve (AUC). The study was registered with the PROSPERO, CRD42022324611.

FINDINGS

Thirty-four eligible studies were identified, of which twenty-eight studies were included in the meta-analysis with a pooled SE of 88% (95%CI: 85-90%), SP of 85% (82-88%), and AUC of 0.93 (0.91-0.95). Analysis for different algorithms revealed a pooled SE of 89% (85-92%) and SP of 88% (82-92%) for machine learning; and a pooled SE of 88% (84-91%) and SP of 84% (80-87%) for deep learning. Acceptable diagnostic performance was demonstrated in subgroup analyses stratified by imaging modalities (Ultrasound, Magnetic Resonance Imaging, or Computed Tomography), sample size (≤300 or >300), AI algorithms versus clinicians, year of publication (before or after 2020), geographical distribution (Asia or non Asia), and the different risk of bias levels (≥3 domain low risk or < 3 domain low risk).

INTERPRETATION

AI algorithms exhibited favorable performance for the diagnosis of OC through medical imaging. More rigorous reporting standards that address specific challenges of AI research could improve future studies.

FUNDING

This work was supported by the Natural Science Foundation of China (No. 82073647 to Q-JW and No. 82103914 to T-TG), LiaoNing Revitalization Talents Program (No. XLYC1907102 to Q-JW), and 345 Talent Project of Shengjing Hospital of China Medical University (No. M0268 to Q-JW and No. M0952 to T-TG).

A Framework for Harmonization of Radiomics Data for Multicenter Studies and Clinical Trials

PURPOSE

Variability in computed tomography images intrinsic to individual scanners limits the application of radiomics in clinical and research settings. The development of reproducible and generalizable radiomics-based models to assess lesions requires harmonization of data. The purpose of this study was to develop, test, and analyze the efficacy of a radiomics data harmonization model.

MATERIALS AND METHODS

Radiomic features from biopsy-proven untreated hepatic metastasis (N = 380) acquired from 167 unique patients with pancreatic, colon, and breast cancers were analyzed. Radiomic features from volume-match 551 samples of normal liver tissue and 188 hepatic cysts were included as references. A novel linear mixed effect model was used to identify effects associated with lesion size, tissue type, and scanner model. Six separate machine learning models were then used to test the effectiveness of radiomic feature harmonization using multivariate analysis.

RESULTS

Proposed model identifies and removes scanner-associated effects while preserving cancer-specific functional dependence of radiomic features on the tumor size. Data harmonization improves the performance of classification models by reducing the scanner-associated variability. For example, the multiclass logistic regression model, LogitBoost, demonstrated the improvement in sensitivity in the range from 15% to 40% for each type of liver metastasis, whereas the overall model accuracy and the kappa coefficient increased by 5% and 8% accordingly.

CONCLUSION

The model removed scanner-associated effects while preserving cancer-specific functional dependence of radiomic features.

The relationship between radiomics and pathomics in Glioblastoma patients: Preliminary results from a cross-scale association study

Glioblastoma multiforme (GBM) typically exhibits substantial intratumoral heterogeneity at both microscopic and radiological resolution scales. Diffusion Weighted Imaging (DWI) and dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) are two functional MRI techniques that are commonly employed in clinic for the assessment of GBM tumor characteristics. This work presents initial results aiming at determining if radiomics features extracted from preoperative ADC maps and post-contrast T1 (T1C) images are associated with pathomic features arising from H&E digitized pathology images. 48 patients from the public available CPTAC-GBM database, for which both radiology and pathology images were available, were involved in the study. 91 radiomics features were extracted from ADC maps and post-contrast T1 images using PyRadiomics. 65 pathomic features were extracted from cell detection measurements from H&E images. Moreover, 91 features were extracted from cell density maps of H&E images at four different resolutions. Radiopathomic associations were evaluated by means of Spearman's correlation (ρ) and factor analysis. p values were adjusted for multiple correlations by using a false discovery rate adjustment. Significant cross-scale associations were identified between pathomics and ADC, both considering features (n = 186, 0.45 < ρ < 0.74 in absolute value) and factors (n = 5, 0.48 < ρ < 0.54 in absolute value). Significant but fewer ρ values were found concerning the association between pathomics and radiomics features (n = 53, 0.5 < ρ < 0.65 in absolute value) and factors (n = 2, ρ = 0.63 and ρ = 0.53 in absolute value). The results of this study suggest that cross-scale associations may exist between digital pathology and ADC and T1C imaging. This can be useful not only to improve the knowledge concerning GBM intratumoral heterogeneity, but also to strengthen the role of radiomics approach and its validation in clinical practice as "virtual biopsy", introducing new insights for omics integration toward a personalized medicine approach.

Radiomics quality score in renal masses: a systematic assessment on current literature

OBJECTIVE

To perform a systematic assessment and analyze the quality of radiomics methodology in current literature in the evaluation of renal masses using the Radiomics Quality Score (RQS) approach.

METHODS

We systematically reviewed recent radiomics literature in renal masses published in PubMed, EMBASE, Elsevier, and Web of Science. Two reviewers blinded by each other's scores evaluated the quality of radiomics methodology in studies published from 2015 to August 2021 using the RQS approach. Owing to the diversity in the imaging modalities and radiomics applications, a meta-analysis could not be performed.

RESULTS

Based on our inclusion/exclusion criteria, a total of 87 published studies were included in our study. The highest RQS was noted in three categories: reporting of clinical utility, gold standard, and feature reduction. The average RQS of the two reviewers ranged from 5 ≤ RQS≤19, with the maximum attainable RQS being 36. Very few (7/87 i.e., 8%) studies received an average RQS that ranged from 17 < RQS≤19, which represents studies with the highest RQS in our study. Many (39/87 i.e., 45%) studies received an average RQS that ranged from 13 < RQS≤15. No significant interreviewer scoring differences were observed.

CONCLUSIONS

We report that the overall scientific quality and reporting of radiomics studies in renal masses is suboptimal, and subsequent studies should bolster current deficiencies to improve reporting of radiomics methodologies.

ADVANCES IN KNOWLEDGE

The RQS approach is a meaningful quantitative scoring system to assess radiomics methodology quality and supports a comprehensive evaluation of the radiomics approach before its incorporation into clinical practice.

Application of Artificial Intelligence Methods for Imaging of Spinal Metastasis

Spinal metastasis is the most common malignant disease of the spine. Recently, major advances in machine learning and artificial intelligence technology have led to their increased use in oncological imaging. The purpose of this study is to review and summarise the present evidence for artificial intelligence applications in the detection, classification and management of spinal metastasis, along with their potential integration into clinical practice. A systematic, detailed search of the main electronic medical databases was undertaken in concordance with the PRISMA guidelines. A total of 30 articles were retrieved from the database and reviewed. Key findings of current AI applications were compiled and summarised. The main clinical applications of AI techniques include image processing, diagnosis, decision support, treatment assistance and prognostic outcomes. In the realm of spinal oncology, artificial intelligence technologies have achieved relatively good performance and hold immense potential to aid clinicians, including enhancing work efficiency and reducing adverse events. Further research is required to validate the clinical performance of the AI tools and facilitate their integration into routine clinical practice.

Impact of image quality on radiomics applications

Radiomics features extracted from medical images have been widely reported to be useful in the patient specific outcome modeling for variety of assessment and prediction purposes. Successful application of radiomics features as imaging biomarkers, however, is dependent on the robustness of the approach to the variation in each step of the modeling workflow. Variation in the input image quality is one of the main sources that impacts the reproducibility of radiomics analysis when a model is applied to broader range of medical imaging data. The quality of medical image is generally affected by both the scanner related factors such as image acquisition/reconstruction settings and the patient related factors such as patient motion. This article aimed to review the published literatures in this field that reported the impact of various imaging factors on the radiomics features through the change in image quality. The literatures were categorized by different imaging modalities and also tabulated based on the imaging parameters and the class of radiomics features included in the study. Strategies for image quality standardization were discussed based on the relevant literatures and recommendations for reducing the impact of image quality variation on the radiomics in multi-institutional clinical trial were summarized at the end of this article.

Radiomic Signatures Associated with CD8+ Tumour-Infiltrating Lymphocytes: A Systematic Review and Quality Assessment Study

The tumour immune microenvironment influences the efficacy of immune checkpoint inhibitors. Within this microenvironment are CD8-expressing tumour-infiltrating lymphocytes (CD8+ TILs), which are an important mediator and marker of anti-tumour response. In practice, the assessment of CD8+ TILs via tissue sampling involves logistical challenges. Radiomics, the high-throughput extraction of features from medical images, may offer a novel and non-invasive alternative. We performed a systematic review of the available literature reporting radiomic signatures associated with CD8+ TILs. We also aimed to evaluate the methodological quality of the identified studies using the Radiomics Quality Score (RQS) tool, and the risk of bias and applicability with the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool. Articles were searched from inception until 31 December 2021, in three electronic databases, and screened against eligibility criteria. Twenty-seven articles were included. A wide variety of cancers have been studied. The reported radiomic signatures were heterogeneous, with very limited reproducibility between studies of the same cancer group. The overall quality of studies was found to be less than desirable (mean RQS = 33.3%), indicating a need for technical maturation. Some potential avenues for further investigation are also discussed.

Assessment of a computed tomography-based radiomics approach for assessing lung function in lung cancer patients

PURPOSE

We aimed to assess radiomics approaches for estimating three pulmonary function test (PFT) results (forced expiratory volume in one second [FEV1], forced vital capacity [FVC], and the ratio of FEV1 to FVC [FEV₁/FVC]) using data extracted from chest computed tomography (CT) images.

METHODS

This retrospective study included 85 lung cancer patients (mean age, 75 years ±8; 69 men) who underwent stereotactic body radiotherapy between 2012 and 2020. Their pretreatment chest breath-hold CT and PFT data before radiotherapy were obtained. A total of 107 radiomics features (Shape: 14, Intensity: 18, Texture: 75) were extracted using two methods: extraction of the lung tissue (<-250 HU) (APPROACH 1), and extraction of small blood vessels and lung tissue (APPROACH 2). The PFT results were estimated using the least absolute shrinkage and selection operator regression. Pearson's correlation coefficients (r) were determined for all PFT results, and the area under the curve (AUC) was calculated for FEV₁/FVC (<70 %). Finally, we compared our approaches with the conventional formula (Conventional).

RESULTS

For the estimated FEV₁/FVC, the Pearson's r were 0.21 (P =.06), 0.69 (P <.01), and 0.73 (P <.01) for Conventional, APPROACH 1, and APPROACH 2, respectively; the AUCs for FEV₁/FVC (<70 %) were 0.67 (95 % confidence interval [CI]: 0.55, 0.79), 0.82 (CI: 0.72, 0.91; P =.047) and 0.86 (CI: 0.78, 0.94; P =.01), respectively.

CONCLUSIONS

The radiomics approach performed better than the conventional equation and may be useful for assessing lung function based on CT images.

Form Factors as Potential Imaging Biomarkers to Differentiate Benign vs. Malignant Lung Lesions on CT Scans

Indeterminate lung nodules detected on CT scans are common findings in clinical practice. Their correct assessment is critical, as early diagnosis of malignancy is crucial to maximise the treatment outcome. In this work, we evaluated the role of form factors as imaging biomarkers to differentiate benign vs. malignant lung lesions on CT scans. We tested a total of three conventional imaging features, six form factors, and two shape features for significant differences between benign and malignant lung lesions on CT scans. The study population consisted of 192 lung nodules from two independent datasets, containing 109 (38 benign, 71 malignant) and 83 (42 benign, 41 malignant) lung lesions, respectively. The standard of reference was either histological evaluation or stability on radiological followup. The statistical significance was determined via the Mann-Whitney U nonparametric test, and the ability of the form factors to discriminate a benign vs. a malignant lesion was assessed through multivariate prediction models based on Support Vector Machines. The univariate analysis returned four form factors (Angelidakis compactness and flatness, Kong flatness, and maximum projection sphericity) that were significantly different between the benign and malignant group in both datasets. In particular, we found that the benign lesions were on average flatter than the malignant ones; conversely, the malignant ones were on average more compact (isotropic) than the benign ones. The multivariate prediction models showed that adding form factors to conventional imaging features improved the prediction accuracy by up to 14.5 pp. We conclude that form factors evaluated on lung nodules on CT scans can improve the differential diagnosis between benign and malignant lesions.

Automatic segmentation of glioblastoma multiform brain tumor in MRI images: Using Deeplabv3+ with pre-trained Resnet18 weights

PURPOSE

To assess the effectiveness of deep learning algorithms in automated segmentation of magnetic resonance brain images for determining the enhanced tumor, the peri-tumoral edema, the necrotic/ non-enhancing tumor, and Normal tissue volumes.

METHODS AND MATERIALS

A new deep neural network algorithm, Deep-Net, was developed for semantic segmentation of the glioblastoma tumors in MR images, using the Deeplabv3+ architecture, and the pre-trained Resnet18 initial weights. The MR image Dataset used for training the network was taken from the BraTS 2020 training set, with the ground truth labels for different tumor subregions manually drawn by a group of expert neuroradiologists. In this work, two multi-modal MRI scans, i.e., T₁ce and FLAIR of 293 patients with high-grade glioma (HGG), were used for deep network training (Deep-Net). The performance of the network was assessed for different hyper-parameters, to obtain the optimum set of parameters. The similarity scores were used for the evaluation of the optimized network.

RESULTS

According to the results of this study, epoch #37 is the optimum epoch giving the best global accuracy (97.53%), and loss function (0.14). The Deep-Net sensitivity in the delineation of the enhanced tumor is more than 90%.

CONCLUSIONS

The results indicate that the Deep-Net was able to segment GBM tumors with high accuracy.