Interobserver variability in tumor contouring affects the use of radiomics to predict mutational status

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.

Assessment of Variabilities in Lung-Contouring Methods on CBCT Preclinical Radiomics Outputs

Simple Summary

This study is the first to evaluate the impact of contouring differences on radiomics analysis in preclinical CBCT scans. We found that the variation in quantitative image readouts was greater between segmentation tools than between observers.

Abstract

Radiomics image analysis has the potential to uncover disease characteristics for the development of predictive signatures and personalised radiotherapy treatment. Inter-observer and inter-software delineation variabilities are known to have downstream effects on radiomics features, reducing the reliability of the analysis. The purpose of this study was to investigate the impact of these variabilities on radiomics outputs from preclinical cone-beam computed tomography (CBCT) scans. Inter-observer variabilities were assessed using manual and semi-automated contours of mouse lungs (n = 16). Inter-software variabilities were determined between two tools (3D Slicer and ITK-SNAP). The contours were compared using Dice similarity coefficient (DSC) scores and the 95th percentile of the Hausdorff distance (HD_95p) metrics. The good reliability of the radiomics outputs was defined using intraclass correlation coefficients (ICC) and their 95% confidence intervals. The median DSC scores were high (0.82–0.94), and the HD_95p metrics were within the submillimetre range for all comparisons. the shape and NGTDM features were impacted the most. Manual contours had the most reliable features (73%), followed by semi-automated (66%) and inter-software (51%) variabilities. From a total of 842 features, 314 robust features overlapped across all contouring methodologies. In addition, our results have a 70% overlap with features identified from clinical inter-observer studies.

Delta radiomics: a systematic review

BACKGROUND

Radiomics can provide quantitative features from medical imaging that can be correlated with various biological features and clinical endpoints. Delta radiomics, on the other hand, consists in the analysis of feature variation at different acquisition time points, usually before and after therapy. The aim of this study was to provide a systematic review of the different delta radiomics approaches.

METHODS

Eligible articles were searched in Embase, PubMed, and ScienceDirect using a search string that included free text and/or Medical Subject Headings (MeSH) with three key search terms: "radiomics", "texture", and "delta". Studies were analysed using QUADAS-2 and the RQS tool.

RESULTS

Forty-eight studies were finally included. The studies were divided into preclinical/methodological (five studies, 10.4%); rectal cancer (six studies, 12.5%); lung cancer (twelve studies, 25%); sarcoma (five studies, 10.4%); prostate cancer (three studies, 6.3%), head and neck cancer (six studies, 12.5%); gastrointestinal malignancies excluding rectum (seven studies, 14.6%), and other disease sites (four studies, 8.3%). The median RQS of all studies was 25% (mean 21% ± 12%), with 13 studies (30.2%) achieving a quality score < 10% and 22 studies (51.2%) < 25%.

CONCLUSIONS

Delta radiomics shows potential benefit for several clinical endpoints in oncology (differential diagnosis, prognosis and prediction of treatment response, and evaluation of side effects). Nevertheless, the studies included in this systematic review suffer from the bias of overall low quality, so that the conclusions are currently heterogeneous, not robust, and not replicable. Further research with prospective and multicentre studies is needed for the clinical validation of delta radiomics approaches.

Impact of Interobserver Variability in Manual Segmentation of Non-Small Cell Lung Cancer (NSCLC) Applying Low-Rank Radiomic Representation on Computed Tomography

Simple Summary

Discovery of predictive and prognostic radiomic features in cancer is currently of great interest to the radiologic and oncologic community. Tumor phenotypic and prognostic information can be obtained by extracting features on tumor segmentations, and it is typically imaging analysts, physician trainees, and attending physicians who provide these labeled datasets for analysis. The potential impact of level and type of specialty training on interobserver variability in manual segmentation of NSCLC was examined. Although there was some variability in segmentation between readers, the subsequently extracted radiomic features were overall well correlated. High fidelity radiomic feature extraction relies on accurate feature extraction from imaging that produce robust prognostic and predictive radiomic NSCLC biomarkers. This study concludes that this goal can be obtained using segmenters of different levels of training and clinical experience.

Abstract

This study tackles interobserver variability with respect to specialty training in manual segmentation of non-small cell lung cancer (NSCLC). Four readers included for segmentation are: a data scientist (BY), a medical student (LS), a radiology trainee (MH), and a specialty-trained radiologist (SK) for a total of 293 patients from two publicly available databases. Sørensen–Dice (SD) coefficients and low rank Pearson correlation coefficients (CC) of 429 radiomics were calculated to assess interobserver variability. Cox proportional hazard (CPH) models and Kaplan-Meier (KM) curves of overall survival (OS) prediction for each dataset were also generated. SD and CC for segmentations demonstrated high similarities, yielding, SD: 0.79 and CC: 0.92 (BY-SK), SD: 0.81 and CC: 0.83 (LS-SK), and SD: 0.84 and CC: 0.91 (MH-SK) in average for both databases, respectively. OS through the maximal CPH model for the two datasets yielded c-statistics of 0.7 (95% CI) and 0.69 (95% CI), while adding radiomic and clinical variables (sex, stage/morphological status, and histology) together. KM curves also showed significant discrimination between high- and low-risk patients (p-value < 0.005). This supports that readers’ level of training and clinical experience may not significantly influence the ability to extract accurate radiomic features for NSCLC on CT. This potentially allows flexibility in the training required to produce robust prognostic imaging biomarkers for potential clinical translation.

Assessment of CT to CBCT contour mapping for radiomic feature analysis in prostate cancer

This study provides a quantitative assessment of the accuracy of a commercially available deformable image registration (DIR) algorithm to automatically generate prostate contours and additionally investigates the robustness of radiomic features to differing contours. Twenty-eight prostate cancer patients enrolled on an institutional review board (IRB) approved protocol were selected. Planning CTs (pCTs) were deformably registered to daily cone-beam CTs (CBCTs) to generate prostate contours (auto contours). The prostate contours were also manually drawn by a physician. Quantitative assessment of deformed versus manually drawn prostate contours on daily CBCT images was performed using Dice similarity coefficient (DSC), mean distance-to-agreement (MDA), difference in center-of-mass position (ΔCM) and difference in volume (ΔVol). Radiomic features from 6 classes were extracted from each contour. Lin’s concordance correlation coefficient (CCC) and mean absolute percent difference in radiomic feature-derived data (mean |%Δ|RF) between auto and manual contours were calculated. The mean (± SD) DSC, MDA, ΔCM and ΔVol between the auto and manual prostate contours were 0.90 ± 0.04, 1.81 ± 0.47 mm, 2.17 ± 1.26 mm and 5.1 ± 4.1% respectively. Of the 1,010 fractions under consideration, 94.8% of DIRs were within TG-132 recommended tolerance. 30 radiomic features had a CCC > 0.90 and 21 had a mean |%∆|RF < 5%. Auto-propagation of prostate contours resulted in nearly 95% of DIRs within tolerance recommendations of TG-132, leading to the majority of features being regarded as acceptably robust. The use of auto contours for radiomic feature analysis is promising but must be done with caution.

Generative models improve radiomics reproducibility in low dose CTs: a simulation study

Radiomics is an active area of research in medical image analysis, however poor reproducibility of radiomics has hampered its application in clinical practice. This issue is especially prominent when radiomic features are calculated from noisy images, such as low dose computed tomography (CT) scans. In this article, we investigate the possibility of improving the reproducibility of radiomic features calculated on noisy CTs by using generative models for denoising. Our work concerns two types of generative models-encoder-decoder network (EDN) and conditional generative adversarial network (CGAN). We then compared their performance against a more traditional 'non-local means' denoising algorithm. We added noise to sinograms of full dose CTs to mimic low dose CTs with two levels of noise: low-noise CT and high-noise CT. Models were trained on high-noise CTs and used to denoise low-noise CTs without re-training. We tested the performance of our model in real data, using a dataset of same-day repeated low dose CTs in order to assess the reproducibility of radiomic features in denoised images. EDN and the CGAN achieved similar improvements on the concordance correlation coefficients (CCC) of radiomic features for low-noise images from 0.87 [95%CI, (0.833, 0.901)] to 0.92 [95%CI, (0.909, 0.935)] and for high-noise images from 0.68 [95%CI, (0.617, 0.745)] to 0.92 [95%CI, (0.909, 0.936)], respectively. The EDN and the CGAN improved the test-retest reliability of radiomic features (mean CCC increased from 0.89 [95%CI, (0.881, 0.914)] to 0.94 [95%CI, (0.927, 0.951)]) based on real low dose CTs. These results show that denoising using EDN and CGANs could be used to improve the reproducibility of radiomic features calculated from noisy CTs. Moreover, images at different noise levels can be denoised to improve the reproducibility using the above models without need for re-training, provided the noise intensity is not excessively greater that of the high-noise CTs. To the authors' knowledge, this is the first effort to improve the reproducibility of radiomic features calculated on low dose CT scans by applying generative models.

Comparative evaluation of conventional and deep learning methods for semi-automated segmentation of pulmonary nodules on CT

Background

Accurate segmentation of pulmonary nodules on computed tomography (CT) scans plays a crucial role in the evaluation and management of patients with suspicion of lung cancer (LC). When performed manually, not only the process requires highly skilled operators, but is also tiresome and time-consuming. To assist the physician in this task several automated and semi-automated methods have been proposed in the literature. In recent years, in particular, the appearance of deep learning has brought about major advances in the field.

Methods

Twenty-four (12 conventional and 12 based on deep learning) semi-automated-'one-click'-methods for segmenting pulmonary nodules on CT were evaluated in this study. The experiments were carried out on two datasets: a proprietary one (383 images from a cohort of 111 patients) and a public one (259 images from a cohort of 100). All the patients had a positive transcript for suspect pulmonary nodules.

Results

The methods based on deep learning clearly outperformed the conventional ones. The best performance [Sørensen-Dice coefficient (DSC)] in the two datasets was, respectively, 0.853 and 0.763 for the deep learning methods, and 0.761 and 0.704 for the traditional ones.

Conclusions

Deep learning is a viable approach for semi-automated segmentation of pulmonary nodules on CT scans.

Quality control of radiomic features using 3D-printed CT phantoms

Purpose: The lack of standardization in quantitative radiomic measures of tumors seen on computed tomography (CT) scans is generally recognized as an unresolved issue. To develop reliable clinical applications, radiomics must be robust across different CT scan modes, protocols, software, and systems. We demonstrate how custom-designed phantoms, imprinted with human-derived patterns, can provide a straightforward approach to validating longitudinally stable radiomic signature values in a clinical setting. Approach: Described herein is a prototype process to design an anatomically informed 3D-printed radiomic phantom. We used a multimaterial, ultra-high-resolution 3D printer with voxel printing capabilities. Multiple tissue regions of interest (ROIs), from four pancreas tumors, one lung tumor, and a liver background, were extracted from digital imaging and communication in medicine (DICOM) CT exam files and were merged together to develop a multipurpose, circular radiomic phantom (18 cm diameter and 4 cm width). The phantom was scanned 30 times using standard clinical CT protocols to test repeatability. Features that have been found to be prognostic for various diseases were then investigated for their repeatability and reproducibility across different CT scan modes. Results: The structural similarity index between the segment used from the patients' DICOM image and the phantom CT scan was 0.71. The coefficient variation for all assessed radiomic features was < 1.0 % across 30 repeat scans of the phantom. The percent deviation (pDV) from the baseline value, which was the mean feature value determined from repeat scans, increased with the application of the lung convolution kernel, changes to the voxel size, and increases in the image noise. Gray level co-occurrence features, contrast, dissimilarity, and entropy were particularly affected by different scan modes, presenting with pDV > ± 15 % . Conclusions: Previously discovered prognostic and popular radiomic features are variable in practice and need to be interpreted with caution or excluded from clinical implementation. Voxel-based 3D printing can reproduce tissue morphology seen on CT exams. We believe that this is a flexible, yet practical, way to design custom phantoms to validate and compare radiomic metrics longitudinally, over time, and across systems.

Systematic Review on the Association of Radiomics with Tumor Biological Endpoints

Radiomics supposes an alternative non-invasive tumor characterization tool, which has experienced increased interest with the advent of more powerful computers and more sophisticated machine learning algorithms. Nonetheless, the incorporation of radiomics in cancer clinical-decision support systems still necessitates a thorough analysis of its relationship with tumor biology. Herein, we present a systematic review focusing on the clinical evidence of radiomics as a surrogate method for tumor molecular profile characterization. An extensive literature review was conducted in PubMed, including papers on radiomics and a selected set of clinically relevant and commonly used tumor molecular markers. We summarized our findings based on different cancer entities, additionally evaluating the effect of different modalities for the prediction of biomarkers at each tumor site. Results suggest the existence of an association between the studied biomarkers and radiomics from different modalities and different tumor sites, even though a larger number of multi-center studies are required to further validate the reported outcomes.

Radiomics Response Signature for Identification of Metastatic Colorectal Cancer Sensitive to Therapies Targeting EGFR Pathway

BACKGROUND

The authors sought to forecast survival and enhance treatment decisions for patients with liver metastatic colorectal cancer by using on-treatment radiomics signature to predict tumor sensitiveness to irinotecan, 5-fluorouracil, and leucovorin (FOLFIRI) alone (F) or in combination with cetuximab (FC).

METHODS

We retrospectively analyzed 667 metastatic colorectal cancer patients treated with F or FC. Computed tomography quality was classified as high (HQ) or standard (SD). Four datasets were created using the nomenclature (treatment) - (quality). Patients were randomly assigned (2:1) to training or validation sets: FCHQ: 78:38, FCSD: 124:62, FHQ: 78:51, FSD: 158:78. Four tumor-imaging biomarkers measured quantitative radiomics changes between standard of care computed tomography scans at baseline and 8 weeks. Using machine learning, the performance of the signature to classify tumors as treatment sensitive or treatment insensitive was trained and validated using receiver operating characteristic (ROC) curves. Hazard ratio and Cox regression models evaluated association with overall survival (OS).

RESULTS

The signature (area under the ROC curve [95% confidence interval (CI)]) used temporal decrease in tumor spatial heterogeneity plus boundary infiltration to successfully predict sensitivity to antiepidermal growth factor receptor therapy (FCHQ: 0.80 [95% CI = 0.69 to 0.94], FCSD: 0.72 [95% CI = 0.59 to 0.83]) but failed with chemotherapy (FHQ: 0.59 [95% CI = 0.44 to 0.72], FSD: 0.55 [95% CI = 0.43 to 0.66]). In cetuximab-containing sets, radiomics signature outperformed existing biomarkers (KRAS-mutational status, and tumor shrinkage by RECIST 1.1) for detection of treatment sensitivity and was strongly associated with OS (two-sided P < .005).

CONCLUSIONS

Radiomics response signature can serve as an intermediate surrogate marker of OS. The signature outperformed known biomarkers in providing an early prediction of treatment sensitivity and could be used to guide cetuximab treatment continuation decisions.

Understanding Sources of Variation to Improve the Reproducibility of Radiomics

Radiomics is the method of choice for investigating the association between cancer imaging phenotype, cancer genotype and clinical outcome prediction in the era of precision medicine. The fast dispersal of this new methodology has benefited from the existing advances of the core technologies involved in radiomics workflow: image acquisition, tumor segmentation, feature extraction and machine learning. However, despite the rapidly increasing body of publications, there is no real clinical use of a developed radiomics signature so far. Reasons are multifaceted. One of the major challenges is the lack of reproducibility and generalizability of the reported radiomics signatures (features and models). Sources of variation exist in each step of the workflow; some are controllable or can be controlled to certain degrees, while others are uncontrollable or even unknown. Insufficient transparency in reporting radiomics studies further prevents translation of the developed radiomics signatures from the bench to the bedside. This review article first addresses sources of variation, which is illustrated using demonstrative examples. Then, it reviews a number of published studies and progresses made to date in the investigation and improvement of feature reproducibility and model performance. Lastly, it discusses potential strategies and practical considerations to reduce feature variability and improve the quality of radiomics study. This review focuses on CT image acquisition, tumor segmentation, quantitative feature extraction, and the disease of lung cancer.

A FDG-PET radiomics signature detects esophageal squamous cell carcinoma patients who do not benefit from chemoradiation

Detection of patients with esophageal squamous cell carcinoma (ESCC) who do not benefit from standard chemoradiation (CRT) is an important medical need. Radiomics using 18-fluorodeoxyglucose (FDG) positron emission tomography (PET) is a promising approach. In this retrospective study of 184 patients with locally advanced ESCC. 152 patients from one center were grouped into a training cohort (n = 100) and an internal validation cohort (n = 52). External validation was performed with 32 patients treated at a second center. Primary endpoint was disease-free survival (DFS), secondary endpoints were overall survival (OS) and local control (LC). FDG-PET radiomics features were selected by Lasso-Cox regression analyses and a separate radiomics signature was calculated for each endpoint. In the training cohort radiomics signatures containing up to four PET derived features were able to identify non-responders in regard of all endpoints (DFS p < 0.001, LC p = 0.003, OS p = 0.001). After successful internal validation of the cutoff values generated by the training cohort for DFS (p = 0.025) and OS (p = 0.002), external validation using these cutoffs was successful for DFS (p = 0.002) but not for the other investigated endpoints. These results suggest that pre-treatment FDG-PET features may be useful to detect patients who do not respond to CRT and could benefit from alternative treatment.

Association between MRI histogram features and treatment response in locally advanced cervical cancer treated by chemoradiotherapy

OBJECTIVE

To examine the associations of histogram features of T2-weighted (T2W) images and apparent diffusion coefficient (ADC) with treatment response in locally advanced cervical cancer (LACC) following concurrent chemoradiotherapy (CCRT).

MATERIALS AND METHODS

Fifty-eight patients who underwent a 4-week CCRT regimen with MRI prior to treatment (pre-CCRT) and after treatment (post-CCRT) were retrospectively analysed. Histogram features were calculated from volumes of interest (VOIs) from one radiologist on T2W images and ADC maps. VOIs from two radiologists were used to assess observer repeatability in delineation and feature values at both time-points with the Dice similarity coefficient (DSC) and intraclass correlation coefficient (ICC). Treatment response was defined as a 90% reduction in tumour volume. Paired Mann-Whitney U tests were used to determine if features changed significantly between examinations. Two-sample Mann-Whitney U tests were used to identify features that were significantly different between response groups. Receiver operating characteristic (ROC) analysis was done on significantly different MRI features between treatment response groups.

RESULTS

Pre-CCRT delineation and feature repeatability were generally good (DSC > 0.700; ICC > 0.750). Post-CCRT repeatability was low (DSC < 0.700; ICC < 0.750), but ADC mean and percentiles retained good ICC scores. All features, except for T2W_Kurtosis, significantly changed between examinations. Post-CCRT ADC₅₀ was the only feature that demonstrated both good observer variability and significant differences between treatment response groups (p = 0.036) and had an AUC of 0.701 with a cut-off of 1.357 × 10^-6 mm²/s.

CONCLUSION

ADC and T2W histogram features could be used to track changes in LACC tumours undergoing CCRT. Post-CCRT ADC₅₀ was associated with treatment response with good observer repeatability.

KEY POINTS

• Pre-treatment tumour delineation and histogram feature values had good observer repeatability, while these were less repeatable at post-treatment. • MRI histogram analysis could be used to track changes in the tumour as it undergoes concurrent chemoradiotherapy. • Post-treatment median ADC was associated with treatment response and had good repeatability.

Radiomics as a personalized medicine tool in lung cancer: Separating the hope from the hype

Radiomics has become a popular image analysis method in the last few years. Its key hypothesis is that medical images harbor biological, prognostic and predictive information that is not revealed upon visual inspection. In contrast to previous work with a priori defined imaging biomarkers, radiomics instead calculates image features at scale and uses statistical methods to identify those most strongly associated to outcome. This builds on years of research into computer aided diagnosis and pattern recognition. While the potential of radiomics to aid personalized medicine is widely recognized, several technical limitations exist which hinder biomarker translation. Aspects of the radiomic workflow lack repeatability or reproducibility under particular circumstances, which is a key requirement for the translation of imaging biomarkers into clinical practice. One of the most commonly studied uses of radiomics is for personalized medicine applications in Non-Small Cell Lung Cancer (NSCLC). In this review, we summarize reported methodological limitations in CT based radiomic analyses together with suggested solutions. We then evaluate the current NSCLC radiomics literature to assess the risk associated with accepting the published conclusions with respect to these limitations. We review different complementary scoring systems and initiatives that can be used to critically appraise data from radiomics studies. Wider awareness should improve the quality of ongoing and future radiomics studies and advance their potential as clinically relevant biomarkers for personalized medicine in patients with NSCLC.

AI-based applications in hybrid imaging: how to build smart and truly multi-parametric decision models for radiomics

INTRODUCTION

The quantitative imaging features (radiomics) that can be obtained from the different modalities of current-generation hybrid imaging can give complementary information with regard to the tumour environment, as they measure different morphologic and functional imaging properties. These multi-parametric image descriptors can be combined with artificial intelligence applications into predictive models. It is now the time for hybrid PET/CT and PET/MRI to take the advantage offered by radiomics to assess the added clinical benefit of using multi-parametric models for the personalized diagnosis and prognosis of different disease phenotypes.

OBJECTIVE

The aim of the paper is to provide an overview of current challenges and available solutions to translate radiomics into hybrid PET-CT and PET-MRI imaging for a smart and truly multi-parametric decision model.

CT Slice Thickness and Convolution Kernel Affect Performance of a Radiomic Model for Predicting EGFR Status in Non-Small Cell Lung Cancer: A Preliminary Study

We evaluated whether the optimal selection of CT reconstruction settings enables the construction of a radiomics model to predict epidermal growth factor receptor (EGFR) mutation status in primary lung adenocarcinoma (LAC) using standard of care CT images. Fifty-one patients (EGFR:wildtype = 23:28) with LACs of clinical stage I/II/IIIA were included in the analysis. The LACs were segmented in four conditions, two slice thicknesses (Thin: 1 mm; Thick: 5 mm) and two convolution kernels (Sharp: B70f/B70s; Smooth: B30f/B31f/B31s), which constituted four groups: (1) Thin-Sharp, (2) Thin-Smooth, (3) Thick-Sharp, and (4) Thick-Smooth. Machine learning algorithms selected and combined 1,695 quantitative image features to build prediction models. The performance of prediction models was assessed by calculating the area under the curve (AUC). The best prediction model yielded AUC (95%CI) = 0.83 (0.68, 0.92) using the Thin-Smooth reconstruction setting. The AUC of models using thick slices was significantly lower than that of thin slices (P < 10-3), whereas the impact of reconstruction kernel was not significant. Our study showed that the optimal prediction of EGFR mutational status in early stage LACs was achieved by using thin CT-scan slices, independently of convolution kernels. Results from the prediction model suggest that tumor heterogeneity is associated with EGFR mutation.

Quantitative imaging of cancer in the postgenomic era: Radio(geno)mics, deep learning, and habitats

Although cancer often is referred to as "a disease of the genes," it is indisputable that the (epi)genetic properties of individual cancer cells are highly variable, even within the same tumor. Hence, preexisting resistant clones will emerge and proliferate after therapeutic selection that targets sensitive clones. Herein, the authors propose that quantitative image analytics, known as "radiomics," can be used to quantify and characterize this heterogeneity. Virtually every patient with cancer is imaged radiologically. Radiomics is predicated on the beliefs that these images reflect underlying pathophysiologies, and that they can be converted into mineable data for improved diagnosis, prognosis, prediction, and therapy monitoring. In the last decade, the radiomics of cancer has grown from a few laboratories to a worldwide enterprise. During this growth, radiomics has established a convention, wherein a large set of annotated image features (1-2000 features) are extracted from segmented regions of interest and used to build classifier models to separate individual patients into their appropriate class (eg, indolent vs aggressive disease). An extension of this conventional radiomics is the application of "deep learning," wherein convolutional neural networks can be used to detect the most informative regions and features without human intervention. A further extension of radiomics involves automatically segmenting informative subregions ("habitats") within tumors, which can be linked to underlying tumor pathophysiology. The goal of the radiomics enterprise is to provide informed decision support for the practice of precision oncology.

Detection of epithelial growth factor receptor (EGFR) mutations on CT images of patients with lung adenocarcinoma using radiomics and/or multi-level residual convolutionary neural networks

Background

We aim to analyze the ability to detect epithelial growth factor receptor (EGFR) mutations on chest CT images of patients with lung adenocarcinoma using radiomics and/or multi-level residual convolutionary neural networks (MCNNs).

Methods

We retrospectively collected 1,010 consecutive patients in Shanghai Chest Hospital from 2013 to 2017, among which 510 patients were EGFR-mutated and 500 patients were wild-type. The patients were randomly divided into a training set (810 patients) and a validation set (200 patients) according to a balanced distribution of clinical features. The CT images and the corresponding EGFR status measured by Amplification Refractory Mutation System (ARMS) method of the patients in the training set were utilized to construct both a radiomics-based model (M_Radiomics) and MCNNs-based model (M_MCNNs). The M_Radiomics and M_MCNNs were combined to build the Model_{Radiomics+MCNNs} (M_{Radiomics+MCNNs}). Clinical data of gender and smoking history constructed the clinical features-based model (M_Clinical). M_Clinical was then added into M_Radiomics, M_MCNNs, and M_{Radiomics+MCNNs} to establish the Model_{Radiomics+Clinical} (M_{Radiomics+Clinical}), the Model_{MCNNs+Clinical} (M_{MCNNs+Clinical}) and the Model_{Radiomics+MCNNs+Clinical} (M_{Radiomics+MCNNs+Clinical}). All the seven models were tested in the validation set to ascertain whether they were competent to detect EGFR mutations. The detection efficiency of each model was also compared in terms of area under the curve (AUC), sensitivity and specificity.

Results

The AUC of the M_Radiomics, M_MCNNs and M_{Radiomics+MCNNs} to predict EGFR mutations was 0.740, 0.810 and 0.811 respectively. The performance of M_MCNNs was better than that of M_Radiomics (P=0.0225). The addition of clinical features did not improve the AUC of the M_Radiomics (P=0.623), the M_MCNNs (P=0.114) and the M_{Radiomics+MCNNs} (P=0.058). The M_{Radiomics+MCNNs+Clinical} demonstrated the highest AUC value of 0.834. The M_MCNNs did not demonstrate any inferiority when compared with the M_{Radiomics+MCNNs} (P=0.742) and the M_{Radiomics+MCNNs+Clinical} (P=0.056).

Conclusions

Both of the M_Radiomics and the M_CNNs could predict EGFR mutations on CT images of patients with lung adenocarcinoma. The M_MCNNs outperformed the M_Radiomics in the detection of EGFR mutations. The combination of these two models, even added with clinical features, is not significantly more efficient than M_MCNNs alone.