Multimodal Deep Learning for Postoperative Liver Cancer Risk Stratification and Intervention

A Multimodal Deep Learning-Driven Study for Perioperative Risk Stratification and Precision Intervention in Hepatocellular Carcinoma Recurrence

This study is for patients with early-stage liver cancer who are planning to have surgery. The goal of this research is to see if a personalized treatment plan, guided by a computer model (an artificial intelligence tool), can help prevent the cancer from coming back after surgery.

First, the computer model will analyze each patient's medical images and health data to predict their personal risk of the cancer returning. Patients whom the model predicts have a high risk of the cancer coming back will be offered a special treatment plan. This plan involves receiving medication (neoadjuvant therapy) before surgery and additional medication (adjuvant therapy) after surgery. The effectiveness of this plan will be compared to the standard approach of surgery alone.

The main goal is to see if this new, personalized plan can better prevent the cancer from returning within 2 years after surgery. The study will also closely monitor the safety of the medications used.

All patients in the study will be followed closely for 2 years with regular scans and check-ups to monitor their health.

Study Overview

• Status: Recruiting
• Conditions: Hepotacellular Carcinoma
• Intervention / Treatment: Combination product: Neoadjuvant HAIC + Lenvatinib + PD-1 Inhibitor; Procedure: Curative Liver Resection; Other: Multimodal AI Risk Stratification

• Study Type: Interventional
• Enrollment (Estimated): 144
• Phase: Phase 2; Phase 1

Contacts and Locations

• Study Contact: Name: Yang WU, M.D.; Phone Number: +8613636076910; Email: 255001907@qq.com

Study Locations

• China
  • Hubei
    • Wuhan, Hubei, China, 438700
      • Recruiting
      • Tongji Hospital
      • Contact: Yang WU, M.D.; Phone Number: +8613636076910; Email: 255001907@qq.com
      • Contact: Email: wgzhang@tjh.tjmu.edu.cn
      • Principal Investigator: Wanguang Zhang, M.D.

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Age and Consent: Patients aged 18-75 years who are able to understand and voluntarily sign an Informed Consent Form.
• Diagnosis: Clinical diagnosis of BCLC stage 0-A hepatocellular carcinoma, confirmed by histopathology or non-invasive imaging criteria per guidelines.
• Surgical Candidacy: Scheduled to undergo curative-intent liver resection.
• Risk Stratification: Predicted as high-risk for aggressive recurrence by the pre-operative multimodal deep learning model (PRE score ≥ 0.5).
• Liver Function: Child-Pugh liver function class A (score ≤ 7).
• Performance Status: ECOG Performance Status of 0 or 1.
• Imaging Requirement: Availability of a standard pre-operative MRI scan (including non-contrast, arterial, portal venous, and delayed phases) performed within 1 month prior to enrollment, with acceptable image quality.
• Follow-up Commitment: Willing and able to comply with the study procedures and scheduled follow-up for at least 2 years.

Exclusion Criteria:

• Pathology: Postoperative pathological confirmation of non-HCC malignancy (e.g., cholangiocarcinoma, combined hepatocellular-cholangiocarcinoma).
• Other Malignancies: History of other active malignancies within the past 5 years, except for appropriately treated carcinoma in situ of the cervix, non-melanoma skin cancer, or other cancers with a very low risk of recurrence.
• Early Mortality/Loss: Death from any cause or loss to follow-up within 90 days after surgery.
• Contraindications to Protocol Therapy: Known hypersensitivity to any component of the neoadjuvant therapy regimen (e.g., oxaliplatin, fluorouracil, PD-1 inhibitors, lenvatinib).
• Severe, uncontrolled medical conditions including but not limited to: Uncontrolled cardiac disease (e.g., NYHA Class III or IV heart failure), Severe renal dysfunction, Uncontrolled hypertension.
• Inability to Participate: Any condition that, in the opinion of the investigator, would compromise the patient's ability to participate in the study or interfere with the evaluation of the study objectives.

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Treatment
• Allocation: Non-Randomized
• Interventional Model: Parallel Assignment
• Masking: None (Open Label)

Number of Arms

2

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: Multimodal AI-guided Neoadjuvant Therapy + Surgery; Participants identified as high-risk for recurrence by the multimodal deep learning (PRE) model receive neoadjuvant therapy prior to curative liver resection. The neoadjuvant regimen consists of hepatic arterial infusion chemotherapy (HAIC) with the mFOLFOX6 regimen, combined with a PD-1 inhibitor and Lenvatinib. This is followed by standard curative liver resection.	Combination product: Neoadjuvant HAIC + Lenvatinib + PD-1 Inhibitor; A combination drug regimen used as neoadjuvant therapy. Includes Hepatic Arterial Infusion Chemotherapy (HAIC) with mFOLFOX6 (Oxaliplatin, Leucovorin, Fluorouracil), oral Lenvatinib, and an intravenous PD-1 inhibitor.; Procedure: Curative Liver Resection; Standard anatomic or non-anatomic liver resection with the intention of achieving complete tumor removal with negative margins. This is the standard surgical procedure for resectable hepatocellular carcinoma; Other: Multimodal AI Risk Stratification; The use of a pre-established deep learning model (PRE/POST model) to analyze preoperative imaging and clinical data to stratify patients' risk of aggressive recurrence. This stratification is used to determine treatment arm assignment.
Active Comparator: Surgery Alone (High-Risk); Participants identified as high-risk for recurrence by the multimodal deep learning (PRE) model proceed directly to standard curative liver resection without receiving neoadjuvant therapy.	Procedure: Curative Liver Resection; Standard anatomic or non-anatomic liver resection with the intention of achieving complete tumor removal with negative margins. This is the standard surgical procedure for resectable hepatocellular carcinoma; Other: Multimodal AI Risk Stratification; The use of a pre-established deep learning model (PRE/POST model) to analyze preoperative imaging and clinical data to stratify patients' risk of aggressive recurrence. This stratification is used to determine treatment arm assignment.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
2-year Recurrence-free Survival (RFS) Rate	The primary outcome is the proportion of patients who remain free of recurrence at 2 years after curative liver resection. Recurrence is defined as radiologically confirmed (via contrast-enhanced CT or MRI according to RECIST 1.1 criteria) appearance of new lesions (intrahepatic or extrahepatic) or death from any cause, whichever occurs first. This outcome will be compared between the intervention arm (AI-guided neoadjuvant therapy + surgery) and the control arm (surgery alone).	2 years post-surgery

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Incidence of Grade 3 or Higher Treatment-Related Adverse Events (TRAEs)	The proportion of patients in the intervention arm who experience at least one Grade 3 or higher adverse event that is judged by the investigator to be related to the neoadjuvant therapy regimen (HAIC + Lenvatinib + PD-1 inhibitor). Severity will be graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (NCI CTCAE) version 5.0.	From the start of neoadjuvant therapy until 30 days after the last dose of neoadjuvant drugs
Overall Survival (OS)	The time from the date of randomization (or enrollment for non-randomized studies) to the date of death from any cause. Patients who are still alive at the time of analysis will be censored at the last known date they were alive.	Up to 3 years post-surgery

Collaborators and Investigators

• Sponsor: Tongji Hospital

Study record dates

Study Major Dates

• Study Start (Actual): October 26, 2025
• Primary Completion (Estimated): June 30, 2027
• Study Completion (Estimated): June 30, 2028

Study Registration Dates

• First Submitted: December 2, 2025
• First Submitted That Met QC Criteria: December 2, 2025
• First Posted (Estimated): December 15, 2025

Study Record Updates

• Last Update Posted (Actual): December 18, 2025
• Last Update Submitted That Met QC Criteria: December 17, 2025
• Last Verified: December 1, 2025

More Information

Terms related to this study

• Keywords: Hepatocellular Carcinoma; Neoadjuvant Therapy; deep learning; recurrence pattern
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Neoplasms by Histologic Type; Digestive System Neoplasms; Digestive System Diseases; Liver Diseases; Neoplasms, Glandular and Epithelial; Adenocarcinoma; Liver Neoplasms; Carcinoma; Carcinoma, Hepatocellular; Antineoplastic Agents, Immunological; Antineoplastic Agents; Molecular Mechanisms of Pharmacological Action; Pharmacologic Actions; Chemical Actions and Uses; Therapeutic Uses; Immune Checkpoint Inhibitors; lenvatinib
• Other Study ID Numbers: TJ-IRB202510021

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No