Diagnosis of rib fractures plays an important role in identifying trauma severity. However, quickly and precisely identifying the rib fractures in a large number of CT images with increasing number of patients is a tough task, which is also subject to the qualification of radiologist. We aim at a clinically applicable automatic system for rib fracture detection and segmentation from CT scans. A total of 7,473 annotated traumatic rib fractures from 900 patients in a single center were enrolled into our dataset, named RibFrac Dataset, which were annotated with a human-in-the-loop labeling procedure. We developed a deep learning model, named FracNet, to detect and segment rib fractures. 720, 60 and 120 patients were randomly split as training cohort, tuning cohort and test cohort, respectively. Free-Response ROC (FROC) analysis was used to evaluate the sensitivity and false positives of the detection performance, and Intersection-over-Union (IoU) and Dice Coefficient (Dice) were used to evaluate the segmentation performance of predicted rib fractures. Observer studies, including independent human-only study and human-collaboration study, were used to benchmark the FracNet with human performance and evaluate its clinical applicability. A annotated subset of RibFrac Dataset, including 420 for training, 60 for tuning and 120 for test, as well as our code for model training and evaluation, was open to research community to facilitate both clinical and engineering research. Our method achieved a detection sensitivity of 92.9% with 5.27 false positives per scan and a segmentation Dice of 71.5%on the test cohort. Human experts achieved much lower false positives per scan, while underperforming the deep neural networks in terms of detection sensitivities with longer time in diagnosis. With human-computer collobration, human experts achieved higher detection sensitivities than human-only or computer-only diagnosis. The proposed FracNet provided increasing detection sensitivity of rib fractures with significantly decreased clinical time consumed, which established a clinically applicable method to assist the radiologist in clinical practice. A full list of funding bodies that contributed to this study can be found in the Acknowledgements section. The funding sources played no role in the study design; collection, analysis, and interpretation of data; writing of the report; or decision to submit the article for publication .

Identification of early-stage pulmonary adenocarcinomas before surgery, especially in cases of subcentimeter cancers, would be clinically important and could provide guidance to clinical decision making. In this study, we developed a deep learning system based on 3D convolutional neural networks and multitask learning, which automatically predicts tumor invasiveness, together with 3D nodule segmentation masks. The system processes a 3D nodule-centered patch of preprocessed CT and learns a deep representation of a given nodule without the need for any additional information. A dataset of 651 nodules with manually segmented voxel-wise masks and pathological labels of atypical adenomatous hyperplasia (AAH), adenocarcinomas in situ (AIS), minimally invasive adenocarcinoma (MIA), and invasive pulmonary adenocarcinoma (IA) was used in this study. We trained and validated our deep learning system on 523 nodules and tested its performance on 128 nodules. An observer study with 2 groups of radiologists, 2 senior and 2 junior, was also investigated. We merged AAH and AIS into one single category AAH-AIS, comprising a 3-category classification in our study. The proposed deep learning system achieved better classification performance than the radiologists; in terms of 3-class weighted average F1 score, the model achieved 63.3% while the radiologists achieved 55.6%, 56.6%, 54.3%, and 51.0%, respectively. These results suggest that deep learning methods improve the yield of discriminative results and hold promise in the CADx application domain, which could help doctors work efficiently and facilitate the application of precision medicine. Significance: Machine learning tools are beginning to be implemented for clinical applications. This study represents an important milestone for this emerging technology, which could improve therapy selection for patients with lung cancer.

To develop a deep learning system based on 3D convolutional neural networks (CNNs), and to automatically predict EGFR‐mutant pulmonary adenocarcinoma in CT images. A dataset of 579 nodules with EGFR mutation status labels of mutant (Mut) or wild‐type (WT) was retrospectively analyzed. A deep learning system, namely 3D DenseNets, was developed to process 3D patches of nodules from CT data, and learn strong representations with supervised end‐to‐end training. The 3D DenseNets were trained with a training subset of 348 nodules and tuned with a development subset of 116 nodules. A strong data augmentation technique, mixup, was used for better generalization. We evaluated our model on a holdout subset of 115 nodules. An independent public dataset of 37 nodules from the cancer imaging archive (TCIA) was also used to test the generalization of our method. Conventional radiomics analysis was also performed for comparison. Our method achieved promising performance on predicting EGFR mutation status, with AUCs of 75.8% and 75.0% for our holdout test set and public test set, respectively. Moreover, strong relations were found between deep learning feature and conventional radiomics, while deep learning worked through an enhanced radiomics manner, that is, deep learned radiomics (DLR), in terms of robustness, compactness and expressiveness. The proposed deep learning system predicts EGFR‐mutant of lung adenocarcinomas in CT images noninvasively and automatically, indicating its potential to help clinical decision‐making by identifying eligible patients of pulmonary adenocarcinoma for EGFR‐targeted therapy.

Objectives: To investigate the value of radiomics based on CT imaging in predicting invasive adenocarcinoma manifesting as pure ground-glass nodules (pGGNs). Methods: This study enrolled 395 pGGNs with histopathology-confirmed benign nodules or adenocarcinoma. A total of 396 radiomic features were extracted from each labeled nodule. A Rad-score was constructed with the least absolute shrinkage and selection operator (LASSO) in the training set. Multivariate logistic regression analysis was conducted to establish the radiographic model and the combined radiographic–radiomics model. The predictive performance was validated by receiver operating characteristic (ROC) curve. Based on the multivariate logistic regression analysis, an individual prediction nomogram was developed and the clinical utility was assessed. Results: Five radiomic features and four radiographic features were selected for predicting the invasive lesions. The combined radiographic–radiomics model (AUC 0.77; 95% CI, 0.69–0.86) performed better than the radiographic model (AUC 0.71; 95% CI, 0.62–0.81) and Rad-score (AUC 0.72; 95% CI, 0.63–0.81) in the validation set. The clinical utility of the individualized prediction nomogram developed using the Rad-score, margin, spiculation, and size was confirmed in the validation set. The decision curve analysis (DCA) indicated that using a model with Rad-score to predict the invasive lesion would be more beneficial than that without Rad-score and the clinical model. Conclusions: The proposed radiomics-based nomogram that incorporated the Rad-score, margin, spiculation, and size may be utilized as a noninvasive biomarker for the assessment of invasive prediction in patients with pGGNs. Key Points: • CT-based radiomics analysis helps invasive prediction manifested as pGGNs. • The combined radiographic–radiomics model may be utilized as a noninvasive biomarker for predicting invasive lesion for pGGNs. • Radiomics-based individual nomogram may serve as a vital decision support tool to identify invasive pGGNs, obviating further workup and blind follow-up.