Artificial Intelligence-Assisted Imaging for Diagnosis and Management of Focal Liver Lesions

An artificial intelligence-based imaging system will be used to assist the diagnosis of focal liver lesions through hierarchical classification. The system will evaluate lesion presence, distinguish benign from malignant lesions, and classify specific malignant and benign lesion types. For patients with independently confirmed hepatocellular carcinoma who undergo interventional therapy, an artificial intelligence-based prognostic model will additionally be evaluated for predicting overall survival and progression-free survival.

Qualified physicians will independently interpret liver imaging examinations according to routine clinical practice without assistance from the study artificial intelligence system. The diagnostic assessment will include lesion detection, differentiation between benign and malignant lesions, and classification of lesion subtypes using established imaging criteria and clinical information.

Detection accuracy is defined as the percentage of participants correctly classified as having or not having focal liver lesions, based on the prespecified reference standard. Accuracy will be calculated as the number of correctly classified participants divided by the total number of participants with an evaluable reference diagnosis, multiplied by 100. Results will be reported separately for the artificial intelligence-assisted and standard diagnostic groups. Higher percentages indicate better diagnostic performance.

The outcome is the percentage of participants whose benign or malignant liver lesion classification matches the reference standard. Binary classification accuracy will be calculated as the number of correctly classified participants (true positives plus true negatives) divided by the total number of evaluable participants with a confirmed benign or malignant focal liver lesion, multiplied by 100. The seven-class diagnostic outputs will be grouped into benign and malignant categories according to a prespecified classification scheme. Results will be reported separately for the artificial intelligence (AI)-assisted and standard diagnostic groups. Higher percentages indicate better diagnostic accuracy.

The outcome is the percentage of participants with correctly classified benign liver lesion subtypes, based on the prespecified reference standard. The analysis will include participants with confirmed hepatic hemangioma or hepatic cyst. Subtype classification accuracy will be calculated as the number of participants whose predicted benign lesion subtype matches the reference diagnosis divided by the total number of evaluable participants with either of these two benign lesions, multiplied by 100. Predictions of an incorrect benign subtype, a malignant category, or no lesion will be counted as incorrect. Results will be reported separately for the artificial intelligence-assisted and standard diagnostic groups. Higher percentages indicate better diagnostic accuracy.

The discriminative performance of the artificial intelligence (AI)-based model for predicting overall survival (OS) in patients with hepatocellular carcinoma (HCC) will be evaluated using Harrell's concordance index (C-index). Overall survival is defined as the time from initiation of the index interventional treatment to death from any cause. Patients who remain alive will be censored at their last known follow-up. The C-index measures the concordance between predicted survival risk and observed survival outcomes, accounting for censored observations. Values range from 0 to 1, with 0.5 indicating discrimination comparable to chance and 1 indicating perfect concordance. Higher values indicate better prognostic discrimination.

The discriminative performance of the artificial intelligence (AI)-based model for predicting progression-free survival (PFS) in patients with hepatocellular carcinoma (HCC) will be evaluated using Harrell's concordance index (C-index). Progression-free survival is defined as the time from initiation of the index interventional treatment to the first documented disease progression or death from any cause, whichever occurs first. Disease progression will be determined according to the prespecified assessment criteria. Patients without progression or death will be censored at their last adequate disease assessment. The C-index measures the concordance between predicted progression risk and observed PFS outcomes, accounting for censored observations. Values range from 0 to 1, with 0.5 indicating discrimination comparable to chance and 1 indicating perfect concordance. Higher values indicate better prognostic discrimination.