EfficientNet-Based System for Detecting EGFR-Mutant Status and Predicting Prognosis of Tyrosine Kinase Inhibitors in Patients with NSCLC.

We aimed to develop and validate a deep learning-based system using pre-therapy computed tomography (CT) images to detect epidermal growth factor receptor (EGFR)-mutant status in patients with non-small cell lung cancer (NSCLC) and predict the prognosis of advanced-stage patients with EGFR mutations treated with EGFR tyrosine kinase inhibitors (TKI). This retrospective, multicenter study included 485 patients with NSCLC from four hospitals. Of them, 339 patients from three centers were included in the training dataset to develop an EfficientNetV2-L-based model (EME) for predicting EGFR-mutant status, and the remaining patients were assigned to an independent test dataset. EME semantic features were extracted to construct an EME-prognostic model to stratify the prognosis of EGFR-mutant NSCLC patients receiving EGFR-TKI. A comparison of EME and radiomics was conducted. Additionally, we included patients from The Cancer Genome Atlas lung adenocarcinoma dataset with both CT images and RNA sequencing data to explore the biological associations between EME score and EGFR-related biological processes. EME obtained an area under the curve (AUC) of 0.907 (95% CI 0.840-0.926) on the test dataset, superior to the radiomics model (P = 0.007). The EME and radiomics fusion model showed better (AUC, 0.941) but not significantly increased performance (P = 0.895) compared with EME. In prognostic stratification, the EME-prognostic model achieved the best performance (C-index, 0.711). Moreover, the EME-prognostic score showed strong associations with biological pathways related to EGFR expression and EGFR-TKI efficacy. EME demonstrated a non-invasive and biologically interpretable approach to predict EGFR status, stratify survival prognosis, and correlate biological pathways in patients with NSCLC.

Ct-based subregional radiomics using hand-crafted and deep learning features for prediction of therapeutic response to anti-PD1 therapy in NSCLC.

PURPOSE: To develop and externally validate subregional radiomics for predicting therapeutic response to anti-PD1 therapy in non-small-cell lung cancer (NSCLC). METHODS: Sixty-six patients from center 1 served as training and internal validation cohorts. Thirty patients from center 2 and thirty patients from center 3 served as external validation 1 and external validation 2 cohorts, respectively. The lesions identified on CT scans were subdivided into two phenotypically consistent subregions by automatic clustering on the patient-level and population-level (denoted as marginal S1 and inner S2). Handcrafted and deep learning-based features were extracted separately from the entire tumor region and subregions, then selected using the intraclass correlation coefficient and least absolute shrinkage and selection operator regression (LASSO). Radiomics signatures (RSs) were built integrating the selected features and correlation coefficients using a logistic regression method. Area under the receiver operating characteristic (ROC) curve (AUC) was calculated to assess the RSs. RESULTS: RSs derived from S1 outperformed those from S2 and the whole tumor region for both handcrafted and deep learning features. The Fusion-RS incorporating the two feature types achieved the best prediction performance in training (AUC = 0.947, 95 % Confidence Interval [CI] 0.905-0.989, SPE = 0.895, SEN = 0.878), internal validation (AUC = 0.875, 95 % CI: 0.782-0.969, SPE = 0.724, SEN = 0.952), external validation 1 (AUC = 0.836, 95 % CI: 0.694-0.977, SPE = 1.000, SEN = 0.533) and external validation 2 (AUC = 0.783, 95 % CI: 0.613-0.953, SPE = 0.765, SEN = 0.692) cohorts. CONCLUSIONS: Subregional radiomics analysis can be useful for predicting therapeutic response to anti-PD1 therapy. The developed Fusion-RS may be considered as a potential non-invasive tool for individual treatment managements.

A general approach for automatic segmentation of pneumonia, pulmonary nodule, and tuberculosis in CT images.

Proposing a general segmentation approach for lung lesions, including pulmonary nodules, pneumonia, and tuberculosis, in CT images will improve efficiency in radiology. However, the performance of generative adversarial networks is hampered by the limited availability of annotated samples and the catastrophic forgetting of the discriminator, whereas the universality of traditional morphology-based methods is insufficient for segmenting diverse lung lesions. A cascaded dual-attention network with a context-aware pyramid feature extraction module was designed to address these challenges. A self-supervised rotation loss was designed to mitigate discriminator forgetting. The proposed model achieved Dice coefficients of 70.92, 73.55, and 68.52% on multi-center pneumonia, lung nodule, and tuberculosis test datasets, respectively. No significant decrease in accuracy was observed (p > 0.10) when a small training sample size was used. The cyclic training of the discriminator was reduced with self-supervised rotation loss (p < 0.01). The proposed approach is promising for segmenting multiple lung lesion types in CT images.

LWMA-Net: Light-weighted morphology attention learning for human embryo grading.

Visual inspection of embryo morphology is routinely used in embryo assessment and selection. However, due to the complexity of morphologies and large inter- and intra-observer variances among embryologists, manual evaluations remain to be subjective and time-consuming. Thus, we proposed a light-weighted morphology attention learning network (LWMA-Net) for automatic assistance on embryo grading. The LWMA-Net integrated a morphology attention module (MAM) to seek the informative features and their locations and a multiscale fusion module (MFM) to increase the features flowing in the model. The LWMA-Net was trained with a primary set of 3599 embryos from 2318 couples that were clinically enrolled between Sep. 2016 and Dec. 2018, and generated area under the receiver operating characteristic curves (AUCs) of 96.88% and 97.58% on 4- and 3-category gradings, respectively. An independent test set comprises 691 embryos from 321 couples between Jan. 2019 and Jan. 2021 were used to test the assisted fertility values on the embryo grading. Five experienced embryologists were invited to regrade the embryos in the independent set with and without the aid of the LWMA-Net three months apart. Embryologists aided by our LWMA-Net significantly improved their grading capabilities with average AUCs improved by 4.98%-5.32% on 4- and 3-category grading tasks, respectively, which suggests good potential of our LWMA-Net on assisted human reproduction.

Development and validation of a deep learning model for survival prognosis of transcatheter arterial chemoembolization in patients with intermediate-stage hepatocellular carcinoma.

PURPOSE: We aimed to develop a deep learning-based approach to evaluate both time-to-progression (TTP) and overall survival (OS) prognosis of transcatheter arterial chemoembolization (TACE) in treatment-naïve patients with intermediate-stage hepatocellular carcinoma (HCC) and compare the approach's performance with those of radiomics and clinical models. METHODS: EfficientNetV2 was used to build a prognosis model for treatment-naïve patients with HCC. Data of 414 intermediate-stage HCC patients from one participant center were collected to construct the training and validation datasets (70%:30%) for TTP prognosis, while data of 129 intermediate-stage HCC patients from another participant center were collected as the test dataset for both TTP and OS prognosis. Three radiomics and three clinical models were then constructed for comparison. RESULTS: Patients with EfficientNetV2-based model score ≤ 0.5 had better TTP than those with higher scores (hazard ratio [HR]: 0.32, 95%CI: 0.22-0.46, P < 0.0001; HR: 0.28, 95%CI: 0.20-0.41, P < 0.0001; and HR: 0.55, 95%CI: 0.36-0.88, P = 0.005 in the training, validation, and test datasets, respectively). Patients with model score ≤ 0.5 had better OS (38.8 months vs 20.9 months, HR: 0.58, 95%CI: 0.37-0.90, P = 0.008). Compared with the radiomics (intra-tumoral and peri-tumoral) and three clinical models, the EfficientNetV2-based model showed better survival prognosis for TACE (P < 0.05) in the test dataset. CONCLUSIONS: The EfficientNetV2-based model enables assessment of both TTP and OS prognosis of TACE in treatment-naïve, intermediate-stage HCC. Patients with lower scores will benefit from TACE. The model can potentially be used by clinicians to improve decision making regarding TACE treatment choices.

A deep learning-based system for survival benefit prediction of tyrosine kinase inhibitors and immune checkpoint inhibitors in stage IV non-small cell lung cancer patients: A multicenter, prognostic study.

Background: For clinical decision making, it is crucial to identify patients with stage IV non-small cell lung cancer (NSCLC) who may benefit from tyrosine kinase inhibitors (TKIs) and immune checkpoint inhibitors (ICIs). In this study, a deep learning-based system was designed and validated using pre-therapy computed tomography (CT) images to predict the survival benefits of EGFR-TKIs and ICIs in stage IV NSCLC patients. Methods: This retrospective study collected data from 570 patients with stage IV EGFR-mutant NSCLC treated with EGFR-TKIs at five institutions between 2010 and 2021 (data of 314 patients were from a previously registered study), and 129 patients with stage IV NSCLC treated with ICIs at three institutions between 2017 and 2021 to build the ICI test dataset. Five-fold cross-validation was applied to divide the EGFR-TKI-treated patients from four institutions into training and internal validation datasets randomly in a ratio of 80%:20%, and the data from another institution was used as an external test dataset. An EfficientNetV2-based survival benefit prognosis (ESBP) system was developed with pre-therapy CT images as the input and the probability score as the output to identify which patients would receive additional survival benefit longer than the median PFS. Its prognostic performance was validated on the ICI test dataset. For diagnosing which patient would receive additional survival benefit, the accuracy of ESBP was compared with the estimations of three radiologists and three oncologists with varying degrees of expertise (two, five, and ten years). Improvements in the clinicians' diagnostic accuracy with ESBP assistance were then quantified. Findings: ESBP achieved positive predictive values of 80·40%, 75·40%, and 77·43% for additional EGFR-TKI survival benefit prediction using the probability score of 0·2 as the threshold on the training, internal validation, and external test datasets, respectively. The higher ESBP score (>0·2) indicated a better prognosis for progression-free survival (hazard ratio: 0·36, 95% CI: 0·19-0·68, p<0·0001) in patients on the external test dataset. Patients with scores >0·2 in the ICI test dataset also showed better survival benefit (hazard ratio: 0·33, 95% CI: 0·18-0·55, p<0·0001). This suggests the potential of ESBP to identify the two subgroups of benefiting patients by decoding the commonalities from pre-therapy CT images (stage IV EGFR-mutant NSCLC patients receiving additional survival benefit from EGFR-TKIs and stage IV NSCLC patients receiving additional survival benefit from ICIs). ESBP assistance improved the diagnostic accuracy of the clinicians with two years of experience from 47·91% to 66·32%, and the clinicians with five years of experience from 53·12% to 61·41%. Interpretation: This study developed and externally validated a preoperative CT image-based deep learning model to predict the survival benefits of EGFR-TKI and ICI therapies in stage IV NSCLC patients, which will facilitate optimized and individualized treatment strategies. Funding: This study received funding from the National Natural Science Foundation of China (82001904, 81930053, and 62027901), and Key-Area Research and Development Program of Guangdong Province (2021B0101420005).

Trends in the application of deep learning networks in medical image analysis: Evolution between 2012 and 2020.

PURPOSE: To evaluate the general rules and future trajectories of deep learning (DL) networks in medical image analysis through bibliometric and hot spot analysis of original articles published between 2012 and 2020. METHODS: Original articles related to DL and medical imaging were retrieved from the PubMed database. For the analysis, data regarding radiological subspecialties; imaging techniques; DL networks; sample size; study purposes, setting, origins and design; statistical analysis; funding sources; authors; and first authors' affiliation was manually extracted from each article. The Bibliographic Item Co-Occurrence Matrix Builder and VOSviewer were used to identify the research topics of the included articles and illustrate the future trajectories of studies. RESULTS: The study included 2685 original articles. The number of publications on DL and medical imaging has increased substantially since 2017, accounting for 97.2% of all included articles. We evaluated the rules of the application of 47 DL networks to eight radiological tasks on 11 human organ sites. Neuroradiology, thorax, and abdomen were frequent research subjects, while thyroid was under-represented. Segmentation and classification tasks were the primary purposes. U-Net, ResNet, and VGG were the most frequently used Convolutional neural network-derived networks. GAN-derived networks were widely developed and applied in 2020, and transfer learning was highlighted in the COVID-19 studies. Brain, prostate, and diabetic retinopathy-related studies were mature research topics in the field. Breast- and lung-related studies were in a stage of rapid development. CONCLUSIONS: This study evaluates the general rules and future trajectories of DL network application in medical image analyses and provides guidance for future studies.

Automatic Lung Nodule Segmentation and Intra-Nodular Heterogeneity Image Generation.

Automatic segmentation of lung nodules on computed tomography (CT) images is challenging owing to the variability of morphology, location, and intensity. In addition, few segmentation methods can capture intra-nodular heterogeneity to assist lung nodule diagnosis. In this study, we propose an end-to-end architecture to perform fully automated segmentation of multiple types of lung nodules and generate intra-nodular heterogeneity images for clinical use. To this end, a hybrid loss is considered by introducing a Faster R-CNN model based on generalized intersection over union loss in generative adversarial network. The Lung Image Database Consortium image collection dataset, comprising 2,635 lung nodules, was combined with 3,200 lung nodules from five hospitals for this study. Compared with manual segmentation by radiologists, the proposed model obtained an average dice coefficient (DC) of 82.05% on the test dataset. Compared with U-net, NoduleNet, nnU-net, and other three models, the proposed method achieved comparable performance on lung nodule segmentation and generated more vivid and valid intra-nodular heterogeneity images, which are beneficial in radiological diagnosis. In an external test of 91 patients from another hospital, the proposed model achieved an average DC of 81.61%. The proposed method effectively addresses the challenges of inevitable human interaction and additional pre-processing procedures in the existing solutions for lung nodule segmentation. In addition, the results show that the intra-nodular heterogeneity images generated by the proposed model are suitable to facilitate lung nodule diagnosis in radiology.

Intratumoral and Peritumoral Analysis of Mammography, Tomosynthesis, and Multiparametric MRI for Predicting Ki-67 Level in Breast Cancer: a Radiomics-Based Study.

PURPOSE: To noninvasively evaluate the use of intratumoral and peritumoral regions from full-field digital mammography (DM), digital breast tomosynthesis (DBT), dynamic contrast-enhanced (DCE), and diffusion-weighted (DW) magnetic resonance imaging (MRI) images separately and combined to predict the Ki-67 level based on radiomics. PROCEDURES: A total of 209 patients with pathologically confirmed breast cancer were consecutively enrolled from September 2017 to March 2021, who underwent DM, DBT, DCE-MRI, and DW MRI scans. Radiomics features were calculated from intratumoral and peritumoral regions in each modality and selected with the least absolute shrinkage and selection operator (LASSO) regression. Radiomics signatures (RSs) were built based on intratumoral, peritumoral, and combined intra- and peritumoral regions. The prediction performance of the RSs was evaluated using the area under the receiver operating characteristic curve (AUC), specificity, and sensitivity as comparison metrics. A nomogram was constructed by integrating the multi-model RS and important clinical predictors and assessed by calibration and decision curve analysis. RESULTS: The combined intra- and peritumoral RSs improved the AUC compared with intra- or peritumoral RSs in each modality. The DCE plus DW MRI yielded higher AUC and specificity but lower sensitivity compared with the DM plus DBT. The nomogram incorporating the multi-model RS, age, and lymph node metastasis status achieved the best prediction performance in the training (AUC, nomogram vs. fusion RS vs. clinical model, 0.922 vs. 0.917 vs. 0.672) and validation (AUCs, nomogram vs. fusion RS vs. clinical model, 0.866 vs. 0.838 vs. 0.661) cohorts. DCA analysis confirmed the potential clinical utility of the nomogram. CONCLUSIONS: Peritumoral regions can provide complementary information to intratumoral regions in mammography and MRI for the prediction of Ki-67 levels. The MRI performed better than mammography in terms of AUC and specificity but weaker in sensitivity. The nomogram has a predictive advantage over each modality and could be a potential tool for predicting Ki-67 levels in breast cancer.

Deep learning predicts epidermal growth factor receptor mutation subtypes in lung adenocarcinoma.

PURPOSE: This study aimed to explore the predictive ability of deep learning (DL) for the common epidermal growth factor receptor (EGFR) mutation subtypes in patients with lung adenocarcinoma. METHODS: A total of 665 patients with lung adenocarcinoma (528/137) were recruited from two different institutions. In the training set, an 18-layer convolutional neural network (CNN) and fivefold cross-validation strategy were used to establish a CNN model. Subsequently, an independent external validation cohort from the other institution was used to evaluate the predictive efficacy of the CNN model. Grad-weighted class activation mapping (Grad-CAM) technology was used for the visual interpretation of the CNN model. In addition, this study also compared the prediction abilities of the radiomics and CNN models. Receiver operating characteristic (ROC) curves, accuracy and precision values, and recall and F1-score were used to evaluate the effectiveness of the CNN model and compare its performance with that of the radiomics model. RESULTS: In the validation set, the micro- and macroaverage values of the area under the ROC curve of the CNN model to identify the three EGFR subtypes were 0.78 and 0.79, respectively. All evaluation indicators of the CNN model were better than those of the radiomics model. CONCLUSIONS: Our study confirmed the potential of DL for predicting the EGFR mutation status in lung adenocarcinoma. The imaging phenotypes of the three mutation subtypes were found to be different, which can provide a basis for choosing more accurate and personalized treatment in patients with lung adenocarcinoma.

Decoding intra-tumoral spatial heterogeneity on radiological images using the Hilbert curve.

BACKGROUND: Current intra-tumoral heterogeneous feature extraction in radiology is limited to the use of a single slice or the region of interest within a few context-associated slices, and the decoding of intra-tumoral spatial heterogeneity using whole tumor samples is rare. We aim to propose a mathematical model of space-filling curve-based spatial correspondence mapping to interpret intra-tumoral spatial locality and heterogeneity. METHODS: A Hilbert curve-based approach was employed to decode and visualize intra-tumoral spatial heterogeneity by expanding the tumor volume to a two-dimensional (2D) matrix in voxels while preserving the spatial locality of the neighboring voxels. The proposed method was validated using three-dimensional (3D) volumes constructed from lung nodules from the LIDC-IDRI dataset, regular axial plane images, and 3D blocks. RESULTS: Dimensionality reduction of the Hilbert volume with a single regular axial plane image showed a sparse and scattered pixel distribution on the corresponding 2D matrix. However, for 3D blocks and lung tumor inside the volume, the dimensionality reduction to the 2D matrix indicated regular and concentrated squares and rectangles. For classification into benign and malignant masses using lung nodules from the LIDC-IDRI dataset, the Inception-V4 indicated that the Hilbert matrix images improved accuracy (85.54% vs. 73.22%, p < 0.001) compared to the original CT images of the test dataset. CONCLUSIONS: Our study indicates that Hilbert curve-based spatial correspondence mapping is promising for decoding intra-tumoral spatial heterogeneity of partial or whole tumor samples on radiological images. This spatial-locality-preserving approach for voxel expansion enables existing radiomics and convolution neural networks to filter structured and spatially correlated high-dimensional intra-tumoral heterogeneity.

Predictive Radiomic Models for the Chemotherapy Response in Non-Small-Cell Lung Cancer based on Computerized-Tomography Images.

The heterogeneity and complexity of non-small cell lung cancer (NSCLC) tumors mean that NSCLC patients at the same stage can have different chemotherapy prognoses. Accurate predictive models could recognize NSCLC patients likely to respond to chemotherapy so that they can be given personalized and effective treatment. We propose to identify predictive imaging biomarkers from pre-treatment CT images and construct a radiomic model that can predict the chemotherapy response in NSCLC. This single-center cohort study included 280 NSCLC patients who received first-line chemotherapy treatment. Non-contrast CT images were taken before and after the chemotherapy, and clinical information were collected. Based on the Response Evaluation Criteria in Solid Tumors and clinical criteria, the responses were classified into two categories: response (n = 145) and progression (n = 135), then all data were divided into two cohorts: training cohort (224 patients) and independent test cohort (56 patients). In total, 1629 features characterizing the tumor phenotype were extracted from a cube containing the tumor lesion cropped from the pre-chemotherapy CT images. After dimensionality reduction, predictive models of the chemotherapy response of NSCLC with different feature selection methods and different machine-learning classifiers (support vector machine, random forest, and logistic regression) were constructed. For the independent test cohort, the predictive model based on a random-forest classifier with 20 radiomic features achieved the best performance, with an accuracy of 85.7% and an area under the receiver operating characteristic curve of 0.941 (95% confidence interval, 0.898-0.982). Of the 20 selected features, four were first-order statistics of image intensity and the others were texture features. For nine features, there were significant differences between the response and progression groups (p < 0.001). In the response group, three features, indicating heterogeneity, were overrepresented and one feature indicating homogeneity was underrepresented. The proposed radiomic model with pre-chemotherapy CT features can predict the chemotherapy response of patients with non-small cell lung cancer. This radiomic model can help to stratify patients with NSCLC, thereby offering the prospect of better treatment.

Multi-classifier-based identification of COVID-19 from chest computed tomography using generalizable and interpretable radiomics features.

PURPOSE: To investigate the efficacy of radiomics in diagnosing patients with coronavirus disease (COVID-19) and other types of viral pneumonia with clinical symptoms and CT signs similar to those of COVID-19. METHODS: Between 18 January 2020 and 20 May 2020, 110 SARS-CoV-2 positive and 108 SARS-CoV-2 negative patients were retrospectively recruited from three hospitals based on the inclusion criteria. Manual segmentation of pneumonia lesions on CT scans was performed by four radiologists. The latest version of Pyradiomics was used for feature extraction. Four classifiers (linear classifier, k-nearest neighbour, least absolute shrinkage and selection operator [LASSO], and random forest) were used to differentiate SARS-CoV-2 positive and SARS-CoV-2 negative patients. Comparison of the performance of the classifiers and radiologists was evaluated by ROC curve and Kappa score. RESULTS: We manually segmented 16,053 CT slices, comprising 32,625 pneumonia lesions, from the CT scans of all patients. Using Pyradiomics, 120 radiomic features were extracted from each image. The key radiomic features screened by different classifiers varied and lead to significant differences in classification accuracy. The LASSO achieved the best performance (sensitivity: 72.2%, specificity: 75.1%, and AUC: 0.81) on the external validation dataset and attained excellent agreement (Kappa score: 0.89) with radiologists (average sensitivity: 75.6%, specificity: 78.2%, and AUC: 0.81). All classifiers indicated that "Original_Firstorder_RootMeanSquared" and "Original_Firstorder_Uniformity" were significant features for this task. CONCLUSIONS: We identified radiomic features that were significantly associated with the classification of COVID-19 pneumonia using multiple classifiers. The quantifiable interpretation of the differences in features between the two groups extends our understanding of CT imaging characteristics of COVID-19 pneumonia.

Decoding COVID-19 pneumonia: comparison of deep learning and radiomics CT image signatures.

PURPOSE: High-dimensional image features that underlie COVID-19 pneumonia remain opaque. We aim to compare feature engineering and deep learning methods to gain insights into the image features that drive CT-based for COVID-19 pneumonia prediction, and uncover CT image features significant for COVID-19 pneumonia from deep learning and radiomics framework. METHODS: A total of 266 patients with COVID-19 and other viral pneumonia with clinical symptoms and CT signs similar to that of COVID-19 during the outbreak were retrospectively collected from three hospitals in China and the USA. All the pneumonia lesions on CT images were manually delineated by four radiologists. One hundred eighty-four patients (n = 93 COVID-19 positive; n = 91 COVID-19 negative; 24,216 pneumonia lesions from 12,001 CT image slices) from two hospitals from China served as discovery cohort for model development. Thirty-two patients (17 COVID-19 positive, 15 COVID-19 negative; 7883 pneumonia lesions from 3799 CT image slices) from a US hospital served as external validation cohort. A bi-directional adversarial network-based framework and PyRadiomics package were used to extract deep learning and radiomics features, respectively. Linear and Lasso classifiers were used to develop models predictive of COVID-19 versus non-COVID-19 viral pneumonia. RESULTS: 120-dimensional deep learning image features and 120-dimensional radiomics features were extracted. Linear and Lasso classifiers identified 32 high-dimensional deep learning image features and 4 radiomics features associated with COVID-19 pneumonia diagnosis (P < 0.0001). Both models achieved sensitivity > 73% and specificity > 75% on external validation cohort with slight superior performance for radiomics Lasso classifier. Human expert diagnostic performance improved (increase by 16.5% and 11.6% in sensitivity and specificity, respectively) when using a combined deep learning-radiomics model. CONCLUSIONS: We uncover specific deep learning and radiomics features to add insight into interpretability of machine learning algorithms and compare deep learning and radiomics models for COVID-19 pneumonia that might serve to augment human diagnostic performance.

Development and Validation of a Machine Learning Model to Explore Tyrosine Kinase Inhibitor Response in Patients With Stage IV EGFR Variant-Positive Non-Small Cell Lung Cancer.

Importance: An end-to-end efficacy evaluation approach for identifying progression risk after epidermal growth factor receptor (EGFR)-tyrosine kinase inhibitor (TKI) therapy in patients with stage IV EGFR variant-positive non-small cell lung cancer (NSCLC) is lacking. Objective: To propose a clinically applicable large-scale bidirectional generative adversarial network for predicting the efficacy of EGFR-TKI therapy in patients with NSCLC. Design, Setting, and Participants: This diagnostic/prognostic study enrolled 465 patients from January 1, 2010, to August 1, 2017, with follow-up from February 1, 2010, to June 1, 2020. A deep learning (DL) semantic signature to predict progression-free survival (PFS) was constructed in the training cohort, validated in 2 external validation and 2 control cohorts, and compared with the radiomics signature. Exposures: An end-to-end bidirectional generative adversarial network framework was designed to predict the progression risk in patients with NSCLC. Main Outcomes and Measures: The primary end point was PFS, considering the time from the initiation of therapy to the date of recurrence, confirmed disease progression, or death. Results: A total of 342 patients with stage IV EGFR variant-positive NSCLC receiving EGFR-TKI therapy met the inclusion criteria. Of these, 145 patients from 2 of the hospitals (n = 117 and 28) formed a training cohort (mean [SD] age, 61 [11] years; 87 [60.0%] female), and the patients from 2 other hospitals comprised 2 external validation cohorts (validation cohort 1: n = 101; mean [SD] age, 57 [12] years; 60 [59.4%] female; and validation cohort 2: n = 96, mean [SD] age, 58 [9] years; 55 [57.3%] female). Fifty-six patients with advanced-stage EGFR variant-positive NSCLC (mean [SD] age, 52 [11] years; 26 [46.4%] female) and 67 patients with advanced-stage EGFR wild-type NSCLC (mean [SD] age, 54 [10] years; 10 [15.0%] female) who received first-line chemotherapy were included. A total of 90 (26%) receiving EGFR-TKI therapy with a high risk of rapid disease progression were identified (median [range] PFS, 7.3 [1.4-32.0] months in the training cohort, 5.0 [0.6-34.6] months in validation cohort 1, and 6.4 [1.8-20.1] months, in validation cohort 2) using the DL semantic signature.The PFS decreased by 36% (hazard ratio, 2.13; 95% CI, 1.30-3.49; P < .001) compared with that in other patients (median [range] PFS, 11.5 [1.5-64.2] months in the training cohort, 10.9 [1.1-50.5] in validation cohort 1, and 8.9 [0.8-40.6] months in validation cohort 2. No significant differences were observed when comparing the PFS of high-risk patients receiving EGFR-TKI therapy with the chemotherapy cohorts (median PFS, 6.9 vs 4.4 months; P = .08). In terms of predicting the tumor progression risk after EGFR-TKI therapy, clinical decisions based on the DL semantic signature led to better survival outcomes than those based on radiomics signature across all risk probabilities by the decision curve analysis. Conclusions and Relevance: This diagnostic/prognostic study provides a clinically applicable approach for identifying patients with stage IV EGFR variant-positive NSCLC who are not likely to benefit from EGFR-TKI therapy. The end-to-end DL-derived semantic features eliminated all manual interventions required while using previous radiomics methods and have a better prognostic performance.

End-to-end automatic differentiation of the coronavirus disease 2019 (COVID-19) from viral pneumonia based on chest CT.

PURPOSE: In the absence of a virus nucleic acid real-time reverse transcriptase-polymerase chain reaction (RT-PCR) test and experienced radiologists, clinical diagnosis is challenging for viral pneumonia with clinical symptoms and CT signs similar to that of coronavirus disease 2019 (COVID-19). We developed an end-to-end automatic differentiation method based on CT images to identify COVID-19 pneumonia patients in real time. METHODS: From January 18 to February 23, 2020, we conducted a retrospective study and enrolled 201 patients from two hospitals in China who underwent chest CT and RT-PCR tests, of which 98 patients tested positive for COVID-19 (118 males and 83 females, with an average age of 42 years). Patient CT images from one hospital were divided among training, validation and test datasets with an 80%:10%:10% ratio. An end-to-end representation learning method using a large-scale bi-directional generative adversarial network (BigBiGAN) architecture was designed to extract semantic features from the CT images. The semantic feature matrix was input for linear classifier construction. Patients from the other hospital were used for external validation. Differentiation accuracy was evaluated using a receiver operating characteristic curve. RESULTS: Based on the 120-dimensional semantic features extracted by BigBiGAN from each image, the linear classifier results indicated that the area under the curve (AUC) in the training, validation and test datasets were 0.979, 0.968 and 0.972, respectively, with an average sensitivity of 92% and specificity of 91%. The AUC for external validation was 0.850, with a sensitivity of 80% and specificity of 75%. Publicly available architecture and computing resources were used throughout the study to ensure reproducibility. CONCLUSION: This study provides an efficient recognition method for coronavirus disease 2019 pneumonia, using an end-to-end design to implement targeted and effective isolation for the containment of this communicable disease.

A review of original articles published in the emerging field of radiomics.

PURPOSE: To determine the characteristics of and trends in research in the emerging field of radiomics through bibliometric and hotspot analyses of relevant original articles published between 2013 and 2018. METHODS: We evaluated 553 original articles concerning radiomics, published in a total of 61 peer-reviewed journals between 2013 and 2018. The following information was retrieved for each article: radiological subspecialty, imaging technique(s), machine learning technique(s), sample size, study setting and design, statistical result(s), study purpose, software used for feature calculation, funding declarations, author number, first author's affiliation, study origin, and journal name. Qualitative and quantitative analyses were performed for the manually extracted data for identification and visualization of the trends in radiomics research. RESULTS: The annual growth rate in the number of published papers was 177.82% (p < 0.001). The characteristics and trends of research hotspots in the field of radiomics were clarified and visualized in this study. It was found that the field of radiomics is at a more mature stage for lung, breast, and prostate cancers than for other sites. Radiomics studies primarily focused on radiological characterization (215) and monitoring (182). Logistic regression and LASSO were the two most commonly used techniques for feature selection. Non-clinical researchers without a medical background dominated radiomics studies (70.52%), the vast majority of which only highlighted positive results (97.80%) while downplaying negative findings. CONCLUSIONS: The reporting of quantifiable knowledge about the characteristics and trajectories of radiomics can inform researchers about the gaps in the field of radiomics and guide its future direction.