Liver Malignancy
Conditions
Keywords
Artificial Intelligence, liver malignancy
Brief summary
This study aims to assess the feasibility of leveraging non-contrast CT and artificial intelligence to detect liver cancer in consecutive real-world patients. To this end, we deploy LEAF in a prospective real-world clinical setting for real-time monitoring, with a particular focus on flagging cases with liver cancer that may be missed by routine clinical workflow.
Detailed description
This prospective real-world trial will be conducted at FAHZU, a high-volume tertiary medical center in mainland China. LEAF will be deployed within the hospital information system through the DAMO Intelligent Medical Imaging interface, allowing it to flag potential liver lesions in real time. Approximately 2500 consecutive patients undergoing non-contrast CT examinations will be enrolled starting in July 2026. All incoming non-contrast chest and abdominal CT scans will be simultaneously reviewed by radiologists in routine clinical workflow and processed by LEAF in real-time. Daily logs of LEAF-positive alerts will be maintained by the research team. A prespecified clinical action committee composed of hepatobiliary surgeons and abdominal radiologists will review the case to assess whether the AI finding warrants communication to the treating physician of these patients. For patients with suspected malignant liver tumors, the committee's consensus on the presence of suspicious lesions will be communicated to their attending physicians, who will then decide whether additional diagnostic assessment is indicated according to standard clinical practice. The standard radiology workflow will not be altered by the study, and LEAF will be evaluated as a risk-stratification and case-flagging tool rather than a replacement for radiologist interpretation.
Interventions
DEVICELEAF(Liver tumor dEtection And classiFication AI)
The LEAF (Liver tumor dEtection And classiFication AI) model will assist in image interpretation. Patients with positive results for liver malignancy while not reported in standard-of-care CT report will be reviewed by a prespecified clinical action committee composed of hepatobiliary surgeons and abdominal radiologists will review the case and decide whether the AI finding warrants communication to the treating physician of these patients. For patients with suspected malignant liver tumors, the committee's consensus on the presence of suspicious lesions will be communicated to their attending physicians, who will then decide whether additional diagnostic assessment is indicated according to standard clinical practice, while remaining blinded to the LEAF results. The standard radiology workflow will not be altered by the study, and LEAF will be evaluated as a risk-stratification and case-flagging tool rather than a replacement for radiologist interpretation.
Sponsors
Zhejiang University
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
DIAGNOSTIC
Masking
NONE
Intervention model description
LEAF is a deep learning-based model that takes non-contrast CT scans as input and performs both tri-class diagnosis (benign, malignant, or non-tumor) and potential lesion localization for liver tumors.
Eligibility
Sex/Gender
ALL
Age
18 Years to 90 Years
Healthy volunteers
Yes
Inclusion criteria
Age range 18 years and above; Underwent non-contrast chest or abdominal CT examination with liver coverage; Patients with an established diagnosis of cirrhosis; Patients with an established diagnosis of extrahepatic cancer.
Exclusion criteria
Patients who have been diagnosed with malignant liver tumor; Patients who underwent liver transplantation; Low quality image, severe artifacts and noise.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Detection accuracy in liver tumor assisted by LEAF (Liver tumor dEtection And classiFication AI) | Within 4 weeks after enrollment | Sensitivity, specificity of liver malignancy identification (defined as liver malignancy vs. liver benign tumor and non-tumor) |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| AI diagnostic performance: patient-level Positive Predictive Value (PPV) and Negative Predictive Value (NPV) of liver malignancy identification | Within 4 weeks after enrollment | — |
| Clinical utility: number of AI-detected and originally overlooked liver malignant lesions | Within 4 weeks after enrollment | recalled and pathologically confirmed |
Countries
China
Outcome results
None listed