A Holistic Weakly Supervised Approach for Liver Tumor Segmentation with Clinical Knowledge-Informed Label Smoothing

Hairong Wang,Lingchao Mao,Zihan Zhang,Jing Li

2024-10-14

Abstract:Liver cancer is a leading cause of mortality worldwide, and accurate CT-based tumor segmentation is essential for diagnosis and treatment. Manual delineation is time-intensive, prone to variability, and highlights the need for reliable automation. While deep learning has shown promise for automated liver segmentation, precise liver tumor segmentation remains challenging due to the heterogeneous nature of tumors, imprecise tumor margins, and limited labeled data. We present a novel holistic weakly supervised framework that integrates clinical knowledge to address these challenges with (1) A knowledge-informed label smoothing technique that leverages clinical data to generate smooth labels, which regularizes model training reducing the risk of overfitting and enhancing model performance; (2) A global and local-view segmentation framework, breaking down the task into two simpler sub-tasks, allowing optimized preprocessing and training for each; and (3) Pre- and post-processing pipelines customized to the challenges of each subtask, which enhances tumor visibility and refines tumor boundaries. We evaluated the proposed method on the HCC-TACE-Seg dataset and showed that these three key components complementarily contribute to the improved performance. Lastly, we prototyped a tool for automated liver tumor segmentation and diagnosis summary generation called MedAssistLiver. The app and code are published at <a class="link-external link-https" href="https://github.com/lingchm/medassist-liver-cancer" rel="external noopener nofollow">this https URL</a>.

Machine Learning,Image and Video Processing

What problem does this paper attempt to address?

The main problem this paper attempts to address is improving the accuracy of automatic liver tumor segmentation in CT images. Specifically, the paper focuses on the following points:

1. **The issue of manual annotation**: Manual annotation of liver tumors is time-consuming and prone to inconsistencies and inter-operator variability, highlighting the need for automated methods.

2. **Challenges of deep learning**: Although deep learning has made some progress in liver segmentation, precise segmentation of liver tumors remains challenging due to factors such as tumor heterogeneity, unclear boundaries, and limited annotated data.

3. **Limitations of existing methods**: While existing deep learning methods perform well in liver segmentation, their performance in liver tumor segmentation is still low, especially when dealing with tumors of different sizes, shapes, and locations.

To address these challenges, the paper proposes a novel holistic weakly supervised framework that integrates clinical knowledge, comprising the following three key components:

1. **Clinical knowledge-based label smoothing technique**: Utilizing clinical data to generate smoothed labels, reducing the risk of overfitting through regularized model training, and enhancing model performance.

2. **Segmentation framework with global and local views**: Decomposing the task into two simpler sub-tasks—global view liver segmentation and local view tumor segmentation—to allow for optimized preprocessing and training for each task.

3. **Customized preprocessing and postprocessing pipeline**: Designing specific preprocessing and postprocessing steps tailored to the challenges of each sub-task to enhance tumor visibility and refine tumor boundaries.

Through these innovations, the paper aims to improve the accuracy and robustness of liver tumor segmentation, thereby providing more reliable support for clinical diagnosis and treatment.