Interventional AI-Human Collaboration for Liver Tumor Diagnosis

AI-human Collaborative Diagnosis of Liver Tumors Using CE-CT

Recent advances in artificial intelligence (AI), particularly deep learning technology, have transformed medical imaging analysis. AI systems have demonstrated diagnostic performance comparable to or exceeding that of expert radiologists in specific tasks. Liver-focused AI diagnostic systems have achieved promising results in multi-center validations; however, these retrospective studies have not yet addressed two critical gaps. First, large-scale prospective trials are required to establish real-world clinical effectiveness. Second, it remains unclear whether AI can be organically embedded into clinical diagnostic workflows to reshape diagnostic and therapeutic pathways, particularly by enhancing the detection and follow-up of hepatic malignancies and ultimately improving patient outcomes.

Study Overview

• Status: Completed
• Conditions: Hepatocellular Carcinoma (HCC); Focal Nodular Hyperplasia; Hepatic Metastasis; Hepatic Hemangioma; Intrahepatic Cholangiocarcinoma (Icc); Cyst
• Intervention / Treatment: Diagnostic test: AI-human collaboration for CE-CTs diagnosis

Detailed Description

This study aims to evaluate the effectiveness of AI-human collaboration in liver tumor diagnosis by embedding real-time AI analysis into conventional multiphasic contrast-enhanced CT (CE-CT) workflows. Specifically, this prospective validation trial will assess diagnostic performance in detecting and characterizing hepatic lesions, particularly malignancies, evaluate the feasibility and efficiency of workflow integration, and determine the potential clinical impact on treatment decision-making and patient management.

• Study Type: Interventional
• Enrollment (Actual): 10333
• Phase: Not Applicable

Contacts and Locations

Study Locations

• China
  • Liaoning
    • Shenyang, Liaoning, China, 110004
      • Shengjing Hospital of China Medical University

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: Yes

Description

Inclusion Criteria:

1. Age range 18 years and above
2. Underwent dynamic contrast-enhanced abdominal CT examination with liver coverage
3. Imaging must include at least three required phases: non-contrast, arterial phase, and venous phase; an delayed phase is optional
4. Complete imaging data that meet AI system analysis requirements.

Exclusion Criteria:

1. History of recent upper-abdominal surgery (within 30 days) or major hepatobiliary-pancreatic surgery affecting liver evaluation (e.g., liver transplantation or Whipple procedure); patients with prior simple cholecystectomy or single-lesion interventional procedures are not excluded
2. History of recent hepatic trauma (within 30 days)
3. Poor image quality or severe noise artifacts (e.g., metal or motion artifacts)
4. Missing required imaging phases (required at least non-contrast, arterial, and venous phases) or inadequate scan range (e.g., lower-abdomen CT such as pelvic or rectal scans not covering the liver)

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Diagnostic
• Allocation: N/A
• Interventional Model: Single Group Assignment
• Masking: None (Open Label)

Number of Arms

1

Arms and Interventions

Participant Group / Arm

Intervention / Treatment

Experimental: AI-human collaboration in CE-CT diagnosis for liver lesions; In the prospective analysis phase, patients undergo routine Multiphasic Contrast-Enhanced Computed Tomography (CE-CT) imaging. The scans are evaluated through two parallel pathways: standard radiologist interpretation (without AI input) and independent AI analysis. When diagnostic discrepancies occur, a senior radiologist or multidisciplinary expert panel reviews the case and provides the definitive diagnosis.

Diagnostic test: AI-human collaboration for CE-CTs diagnosis; The system automatically processes all eligible same-day scans and generates results for review the following day. To maintain efficient AI-human collaboration while preserving the standard clinical workflow, the conventional radiological interpretation process remains unchanged (first-line radiologists provide initial reports followed by senior radiologists' review). A dedicated senior radiologist then evaluates any discordances between AI findings and primary radiological report. For complex cases, the review process escalates to a consensus review panel (i.e., pre-designated senior radiologists, Multidisciplinary Team (MDT)). The MDT can recommend clinical interventions including follow-up (e.g., additional imaging examinations, active surveillance), surgical procedures, or adjustments to adjuvant therapy (initiation or modification of treatment regimens). All discordant cases and their outcomes are systematically documented for longitudinal tracking and follow-up analysis.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Diagnostic accuracy of the AI System for malignancy diagnosis	Measures the patient-level diagnostic accuracy of the AI system for differentiating malignant vs. non-malignant lesions. The primary metric is the Area under the Receiver Operating Characteristic Curve (AUC). The primary analysis will test the one-sided superiority hypothesis H1: AUC > 0.90 against H0: AUC <= 0.90. The trial will be considered successful if the lower bound of the 95% Confidence Interval (CI) for the AUC is greater than 0.90.	Up to 90 days

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Secondary diagnostic performance	Measures the patient-level diagnostic performance of the AI system for malignant versus non-malignant classification. Metrics include sensitivity, specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV). These will be calculated from the continuous probability score using a fixed operating point prior to prospective analysis.	Up to 90 days
Lesion screening performance	Measures the patient-level screening ability of the AI system to distinguish patients with any lesion from those with no lesions. This is a binary classification task (AUC) comparing lesion patient (malignant or benign) versus no lesion (normal liver or diffuse disease only).	Up to 90 days
Detection discordance	Measures the number of FLLs identified by the AI-human collaborative workflow that were overlooked by the initial radiologist report. An overlooked lesion is defined as an event meeting all three criteria: (1) detected by the AI system; (2) not described in the initial radiological report; (3) confirmed as a true lesion by senior radiologist/MDT re-review.	Up to 90 days
Amended radiological report	Measures the number of formal addenda issued to finalized radiology reports. An amended report is defined as a formal addendum that explicitly corrects a diagnosis or adds a previously missed finding based on the AI-human collaborative review.	Up to 90 days

Collaborators and Investigators

• Sponsor: Shengjing Hospital
• Investigators: Principal Investigator: Yu Shi, MD PhD, Shengjing Hospital

Publications and helpful links

General Publications

• Cao K, Xia Y, Yao J, Han X, Lambert L, Zhang T, Tang W, Jin G, Jiang H, Fang X, Nogues I, Li X, Guo W, Wang Y, Fang W, Qiu M, Hou Y, Kovarnik T, Vocka M, Lu Y, Chen Y, Chen X, Liu Z, Zhou J, Xie C, Zhang R, Lu H, Hager GD, Yuille AL, Lu L, Shao C, Shi Y, Zhang Q, Liang T, Zhang L, Lu J. Large-scale pancreatic cancer detection via non-contrast CT and deep learning. Nat Med. 2023 Dec;29(12):3033-3043. doi: 10.1038/s41591-023-02640-w. Epub 2023 Nov 20.
• Ding W, Meng Y, Ma J, Pang C, Wu J, Tian J, Yu J, Liang P, Wang K. Contrast-enhanced ultrasound-based AI model for multi-classification of focal liver lesions. J Hepatol. 2025 Aug;83(2):426-439. doi: 10.1016/j.jhep.2025.01.011. Epub 2025 Jan 21.
• Ying H, Liu X, Zhang M, Ren Y, Zhen S, Wang X, Liu B, Hu P, Duan L, Cai M, Jiang M, Cheng X, Gong X, Jiang H, Jiang J, Zheng J, Zhu K, Zhou W, Lu B, Zhou H, Shen Y, Du J, Ying M, Hong Q, Mo J, Li J, Ye G, Zhang S, Hu H, Sun J, Liu H, Li Y, Xu X, Bai H, Wang S, Cheng X, Xu X, Jiao L, Yu R, Lau WY, Yu Y, Cai X. A multicenter clinical AI system study for detection and diagnosis of focal liver lesions. Nat Commun. 2024 Feb 7;15(1):1131. doi: 10.1038/s41467-024-45325-9.

Study record dates

Study Major Dates

• Study Start (Actual): September 1, 2025
• Primary Completion (Actual): October 29, 2025
• Study Completion (Actual): November 7, 2025

Study Registration Dates

• First Submitted: August 26, 2025
• First Submitted That Met QC Criteria: August 26, 2025
• First Posted (Estimated): September 4, 2025

Study Record Updates

• Last Update Posted (Actual): November 18, 2025
• Last Update Submitted That Met QC Criteria: November 14, 2025
• Last Verified: November 1, 2025

More Information

Terms related to this study

• Keywords: Dynamic Contrast-Enhanced CT; Focal Liver Lesions
• Additional Relevant MeSH Terms: Neoplasms by Site; Neoplasms; Pathological Conditions, Anatomical; Neoplasms by Histologic Type; Digestive System Neoplasms; Digestive System Diseases; Liver Diseases; Neoplasms, Glandular and Epithelial; Adenocarcinoma; Liver Neoplasms; Carcinoma; Pathological Conditions, Signs and Symptoms; Carcinoma, Hepatocellular; Cholangiocarcinoma; Cysts; Cirrhosis, Familial, with Pulmonary Hypertension; Focal Nodular Hyperplasia
• Other Study ID Numbers: SH-CMU-FLL-Intervention

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: YES
• IPD Plan Description: We plan to share IPD related to abdominal dynamic-contrast enhanced CT scans and clinical outcomes for hepatic tumor diagnosis.
• IPD Sharing Supporting Information Type: STUDY_PROTOCOL

Drug and device information, study documents

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