LiMT: A Multi-task Liver Image Benchmark Dataset

TL;DR

LiMT introduces a public 3D arterial-phase CT dataset that unifies liver and tumor segmentation, lesion detection, and multi-label lesion classification across 150 cases with four lesion types plus normal. The dataset integrates anatomy and lesion-type annotations to enable cross-task learning and reduce training heterogeneity, accompanied by baseline benchmarks for segmentation, detection, and classification. Experimental results show strong performance for standard segmentation models (e.g., nn-UNet), notable gains from multi-task learning, and competitive detection with attention-based methods, underscoring the dataset’s value as a cross-task benchmark. LiMT’s public availability and planned expansion aim to advance robust liver CAD research, multi-center generalization, and richer analytic capabilities for liver disease assessment.

Abstract

Computer-aided diagnosis (CAD) technology can assist clinicians in evaluating liver lesions and intervening with treatment in time. Although CAD technology has advanced in recent years, the application scope of existing datasets remains relatively limited, typically supporting only single tasks, which has somewhat constrained the development of CAD technology. To address the above limitation, in this paper, we construct a multi-task liver dataset (LiMT) used for liver and tumor segmentation, multi-label lesion classification, and lesion detection based on arterial phase-enhanced computed tomography (CT), potentially providing an exploratory solution that is able to explore the correlation between tasks and does not need to worry about the heterogeneity between task-specific datasets during training. The dataset includes CT volumes from 150 different cases, comprising four types of liver diseases as well as normal cases. Each volume has been carefully annotated and calibrated by experienced clinicians. This public multi-task dataset may become a valuable resource for the medical imaging research community in the future. In addition, this paper not only provides relevant baseline experimental results but also reviews existing datasets and methods related to liver-related tasks. Our dataset is available at https://drive.google.com/drive/folders/1l9HRK13uaOQTNShf5pwgSz3OTanWjkag?usp=sharing.

Figures (6)

• Figure 1: The distribution of the lesion number and volume of LiTS dataset and LiMT dataset. In the figure, hepatocellular carcinoma is denoted as HCC, hepatic metastases as HM, cavernous hemangioma of liver as CHL, and liver cyst as HC.
• Figure 2: Samples of four liver lesions in the dataset.
• Figure 3: The flowchart of labeling the LiMT dataset.
• Figure 4: Images and their corresponding annotations (The liver is annotated in red. Hepatocellular carcinoma, hepatic metastases, hepatic cyst, and cavernous hemangioma of liver are annotated in purple, blue, green, and yellow, respectively.)
• Figure 5: Statistical distributions of the dataset. In the figure, hepatocellular carcinoma is denoted as HCC, hepatic metastases as HM, cavernous hemangioma of liver as CHL, and liver cyst as HC. (a) The number of liver disease slices. (b) Distribution of diagnosis results. (c) Distribution of gender. (d) Distribution of age groups. (e) The proportion of four types of liver diseases in each age grade. (f) Number of liver cases collected by year. (g) Intensity distribution of each liver disease. (h) Area distribution of each liver disease in each slice. (i) Volume distribution of each liver disease.