AI-Based Prediction of Liver Metastasis in Colorectal Cancer (A Retrospective Study)

Tongji Hospital1,500 enrolled

Overview

This multicenter, retrospective study aims to develop and validate a multimodal deep learning model for predicting the risk of metachronous liver metastasis in patients with stage I-III colorectal cancer following curative resection. The model will integrate preoperative contrast-enhanced CT imaging, digitized histopathological whole-slide images, and standard clinical-pathological data. The primary objective is to assess the model's discriminatory performance, measured by the area under the receiver operating characteristic curve (AUC), and to compare its predictive accuracy against traditional prognostic factors such as TNM staging and serum carcinoembryonic antigen levels. This research utilizes existing archival data; no direct patient contact or intervention is involved. The ultimate goal is to provide a robust, data-driven tool for improved risk stratification, which could potentially guide personalized surveillance strategies and adjuvant therapy decisions in the future.

Study Type

OBSERVATIONAL

Enrollment

1,500

Conditions

Colorectal Cancer Liver Metastases (CRLM)

Interventions

Multimodal Deep Learning Model Analysis

This is a non-interventional study. The primary study procedure is the application of a multimodal deep learning model to retrospectively analyze existing clinical data (contrast-enhanced CT images, digitized pathology slides, and structured clinical variables) for the purpose of predicting the risk of metachronous liver metastasis. No therapeutic or diagnostic interventions are administered to participants as part of this research protocol.

Eligibility

Sex: ALLMin age: 18 YearsMax age: 75 Years

Medical Language ↔ Plain English

Inclusion Criteria: * Age 18-75 years, any gender. * Histologically confirmed primary colon or rectal adenocarcinoma. * Underwent curative radical resection (R0 resection) for colorectal cancer. * Preoperative contrast-enhanced abdominal/pelvic CT scan performed within 1 month before surgery, with acceptable image quality. * No evidence of distant metastasis (including synchronous liver metastasis) on preoperative or intraoperative exploration. Exclusion Criteria: * History of other malignant tumors. * Previous history of liver surgery or liver transplantation. * Missing clinical, imaging, or pathological data required for the study. * Death within the perioperative period (within 30 days after surgery). * Lack of regular follow-up information.

Outcomes

Primary Outcomes

Area Under the Receiver Operating Characteristic Curve (AUC)

The discriminatory performance of the multimodal deep learning model for predicting the 3-year risk of metachronous liver metastasis. The model integrates preoperative contrast-enhanced CT images, digitized whole-slide pathology images, and clinical data. The AUC will be calculated on the held-out independent test set. The assessment is based on data collected from the date of curative surgery (baseline) to the date of first imaging-confirmed liver metastasis or last follow-up.

Time frame: up to 3 years

Secondary Outcomes

Liver Metastasis-Free Survival (LMFS) by Risk Group

The difference in liver metastasis-free survival between the high-risk and low-risk groups as stratified by the multimodal model. LMFS is defined as the time from surgery to the first radiological diagnosis of liver metastasis. From the date of surgery until the date of first documented liver metastasis or last follow-up.