A novel deep learning model may accurately predict the risk for overall survival (OS) based on computed tomography (CT) images.
Improved risk prediction of OS in gastric cancer is needed in order to guide personalized treatment. Deep learning models may serve as a tool for risk prediction.
A group of Chinese researchers designed a deep learning model that classifies patients with gastric cancer into low- and high-risk groups for OS based on CT images. Researchers retrospectively sampled 640 patients from three independent cancer centers. Patients from center 1 and center 2 were divided into a training cohort (n = 518) and patients from center 3 served as an external validation cohort (n = 122).
A deep learning model based on the architecture of residual convolutional neural network was developed. Researchers augmented the size of the training dataset by image transformations to avoid overfitting, and they also developed radiomics and clinical models for comparison.
Performance of the three models were comprehensively assessed. Results of the analysis were published in Radiotherapy and Oncology (online June 12, 2020; doi:10.1016/j.radonc.2020.06.010).
In total, 518 patients were prepared by data augmentation and inputted into the deep learning model. The model significantly classified patients into high-risk and low-risk groups in the training cohort (P < .001; concordance index, 0.82; HR, 9.79) as well as in the external validation cohort (P < .001; concordance index, 0.78; HR, 11.76).
The radiomics model was developed with 24 selected features and the clinical model was developed with three significant clinical variables (P < .05).
Researchers found that the deep learning model had the best performance for risk prediction of OS compared with the clinical and radiomic models (concordance index, 0.82 vs 0.73 vs 0.66, respectively). External validation yielded similar results (concordance index, 0.78 vs 0.71 vs 0.72, respectively).
In their concluding remarks, authors of the study wrote that the deep learning model is a powerful model for risk assessment of OS in patients with gastric cancer, adding that it could potentially serve as an individualized recommender for decision-making in these patients.—Zachary Bessette
