This study aims to develop a comprehensive artificial intelligence model system integrating preoperative multimodal data (CT/MRI imaging, clinical laboratory data, and radiology report text) to achieve two core objectives. First, to develop a multimodal fusion diagnostic model for non-invasive and accurate preoperative differentiation of liver cancer subtypes, including distinguishing benign from malignant lesions and differentiating hepatocellular carcinoma from intrahepatic cholangiocarcinoma. Second, to develop a prognostic prediction model for patients with confirmed liver cancer undergoing radical surgery to assess postoperative progression-free survival and overall survival. This is a multicenter retrospective cohort study with an anticipated sample size of ≥600 patients. Model performance will be evaluated using AUC, accuracy, sensitivity, specificity, C-index, and calibration curves. Subgroup analysis will be conducted based on whether patients received neoadjuvant therapy.
Eligibility
Sex
ALL
Min age
18 Years
Max age
80 Years
Healthy volunteers
No
Inclusion Criteria:
-Diagnostic Model Cohort:
* Age ≥18 years
* Underwent preoperative contrast-enhanced CT or MRI for clinically suspected liver space-occupying lesion
* Have complete preoperative clinical laboratory data
* Have complete original CT/MRI imaging data and radiology reports
* Have definite pathological diagnosis from surgery or biopsy as gold standard
Prognostic Prediction Model Cohort (selected from diagnostic cohort):
* Meet all diagnostic cohort inclusion criteria
* Pathologically confirmed liver cancer
* Underwent radical hepatectomy
* Have complete preoperative multimodal data (CT/MRI imaging, clinical laboratory data, radiology reports)
* Have complete postoperative follow-up data to determine progression-free survival and overall survival endpoints and time (minimum follow-up of 24 months)
Exclusion Criteria:
* · Key clinical, imaging, or pathological data severely missing or incomplete
* Preoperative CT or MRI images of poor quality or missing sequences, unable to perform reliable image analysis
* Prior local treatment for the target liver lesion, unless clearly recorded as neoadjuvant therapy before surgery
* Concurrent other malignant tumors
* Lost to follow-up or follow-up data cannot meet endpoint determination requirements
Primary outcome measure(s)
Diagnostic Accuracy of the Multimodal AI Model for Liver Lesion Classification — At the time of initial diagnosis The diagnostic performance of the multimodal AI model in differentiating benign from malignant liver lesions and distinguishing hepatocellular carcinoma from intrahepatic cholangiocarcinoma, evaluated using pathology results as the gold standard. Performance metrics include area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity.
Prognostic Performance of the Multimodal AI Model for Postoperative Survival Prediction — minimum follow-up of 24 months The prognostic performance of the multimodal AI model in predicting postoperative progression-free survival (PFS) and overall survival (OS) in patients with pathologically confirmed liver cancer who underwent radical hepatectomy. Performance metric includes the concordance index (C-index). Calibration curves are also assessed.
Trial sites (1)
Facility
City
Region
Status
Guangxi Medical University First Affiliated Hospital
This page summarises publicly available registry data for informational purposes — not medical advice. Eligibility is determined by each study team; patients should discuss participation with their clinician.
We use cookies to analyse site traffic and improve your experience. With your consent, we may also use cookies for advertising. You can change your choice at any time on our Cookie Policy page. See also our Privacy Policy.