Development and Evaluation of an AI-Based Computer-Aided System Using Preoperative CT Images to Predict Microvascular Invasion in Hepatocellular Carcinoma

What this Trial Is About

Hepatocellular carcinoma HCC is a common and serious liver cancer in China with a high death rate. Early recurrence of the cancer is closely linked to how aggressive the tumor is. Microvascular invasion MVI, which means cancer cells are found in small blood vessel branches near the tumor, is an important sign of aggressive disease and is linked to early return of cancer after surgery. Currently, MVI can only be definitively diagnosed after surgery, and there are no reliable methods to predict it before surgery. This research aims to create and test an advanced computer-aided diagnosis CAD system using artificial intelligence and detailed CT images taken before surgery to predict MVI in patients with HCC. The CAD system combines multiple types of imaging features including deep learning-extracted patterns, specific tumor characteristics like shape and texture, and expert-defined factors such as tumor edge clarity and blood vessel relationships. It uses a hybrid model combining convolutional neural networks, Transformer modules, and graph neural networks to analyze tumor details and surrounding areas. The system also provides confidence levels and visual outputs to help doctors understand and trust the predictions. This study will observe its performance across multiple hospitals by comparing the systems predictions with actual pathology results after surgery. Participants are adults aged 18 and older who are scheduled for surgical removal of HCC and have preoperative multiphase CT scans and pathological evaluation data available. Researchers will collect imaging, clinical, and pathology data to evaluate how accurately the CAD system predicts MVI compared to standard pathology. The main measures include the systems accuracy and the area under the receiver operating characteristic curve AUC within one month after surgery. Additional measures include sensitivity, specificity, and calibration. The study will also examine how the systems predictions might influence surgical decision-making and risk assessment in real-world practice.

Computer-Aided Diagnosis for Hepatocellular Carcinoma Microvascular Invasion