LIDIA: Precise Liver Tumor Diagnosis on Multi-Phase Contrast-Enhanced CT via Iterative Fusion and Asymmetric Contrastive Learning

TL;DR

LIDIA tackles accurate liver tumor diagnosis from multi-phase DCE-CT under real-world conditions where some phases may be missing and tumor types are highly heterogeneous. It extends Mask2Former with an Iterative Fusion Module to fuse variable-phase CT data and with Asymmetric Contrastive Learning to better separate common and rare tumor classes, complemented by LiverMax for patient-level predictions. On a large-scale dataset of 1,921 patients and 8,138 lesions, LIDIA achieves a mean patient-wise AUC of 0.936 (AUC-8) and strong external generalization (AUC 0.893), outperforming baselines and ablations confirm the contributions of IFM and ACL. The method demonstrates practical clinical impact by robustly leveraging available phase information and improving discrimination across heterogeneous lesion types, including rare ones.

Abstract

The early detection and precise diagnosis of liver tumors are tasks of critical clinical value, yet they pose significant challenges due to the high heterogeneity and variability of liver tumors. In this work, a precise LIver tumor DIAgnosis network on multi-phase contrast-enhance CT, named LIDIA, is proposed for real-world scenario. To fully utilize all available phases in contrast-enhanced CT, LIDIA first employs the iterative fusion module to aggregate variable numbers of image phases, thereby capturing the features of lesions at different phases for better tumor diagnosis. To effectively mitigate the high heterogeneity problem of liver tumors, LIDIA incorporates asymmetric contrastive learning to enhance the discriminability between different classes. To evaluate our method, we constructed a large-scale dataset comprising 1,921 patients and 8,138 lesions. LIDIA has achieved an average AUC of 93.6% across eight different types of lesions, demonstrating its effectiveness. Besides, LIDIA also demonstrated strong generalizability with an average AUC of 89.3% when tested on an external cohort of 828 patients.

Figures (4)

• Figure 1: Illustration of the overall framework of LIDIA.
• Figure 2: Illustration of AUC for all classes on an external cohort.
• Figure 3: Qualitative examples of lesion segmentation and classification in DCE-CT using different methods.
• Figure 4: Confusion matrix of lesion-level tumor classification in DCE-CT for recalled samples.