Improving B-Mode Ultrasound  Diagnostic Performance for Focal Liver Lesions Using Deep Learning: A Prospective Multicentre Study

Abstract

Background: The diagnosis performance of B-mode ultrasound (US) for focal liver lesions (FLLs) is relatively limited. The aim of this study was to develop a deep convolutional neural network of US (DCNN-US) for aiding radiologists in classification of malignant from benign FLLs with high sensitivity and specificity. Methods: This prospective multicentre study was conducted in 13 hospitals with different levels in China. Patients who had non-cystic FLLs with pathological results were enrolled. The FLLs from 11 hospitals were randomly divided into training and internal validations (IV) cohorts with a 4:1 ratio for developing and evaluating DCNN-US. Diagnostic performance of the model was verified using external validation (EV) cohort from another two hospitals. Three models were developed by using lesion (ModelLesion, ModelL), liver background images (Model Lesion+Background, ModelL B) and clinical factors (ModelLesion+Background +clinic, ModelL BC) to diagnose FLLs, respectively. The diagnosis value of DCNN-US was compared with that of contrast enhanced computed tomography (CT)/magnetic resonance image (MRI) and 236 radiologists, respectively. Findings: We enrolled 24,343 US images of 2,143 patients to develop and test DCNN-US model. The diagnostic area under the curve (AUC) values of ModelL B increased 6% than ModelL, and Model L BC increased nearly 11% than ModelL B. The AUC of ModelLBC for FLLs were 0.925 (95% CI: 0.886–0.963), and 0.924 (95% CI: 0.889–0.959) in the IV and EV cohorts, respectively. The diagnostic sensitivity and specificity of ModelL BC were superior to 15-year skilled radiologists (86.5% vs 76.1%, and 85.5% vs 75.9%, respectively) (both p<0.01). Accuracy of ModelL BC was comparable to that of contrast enhanced CT (both 84.7%) but inferior to contrast enhanced MRI (87.9%). Interpretation: DCNN-US with high sensitivity and specificity in diagnosing FLLs shows its potential to assist less-experienced radiologists in improving their performance and lowering their dependence on sectional imaging in liver cancer diagnosis. Trial Registration: ClinicalTrials.gov (NCT03871140). Funding Statement: The National Social Science Foundation, the National Key R&D Program, the National Scientific Foundation Committee of Beijing, Ministry of Science and Technology, the Strategic Priority Research Program of China, Chinese Academy of Sciences and Beijing Municipal Science & Technology Commission. Declaration of Interests: No conflicts of interest. Ethics Approval Statement: This prospective multicentre study was approved by the ethics committee of each centre.