Holistic multitask regression network for multiapplication shape regression segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Machine Learning

Background:

• Clinical image analysis faces limitations in identifying multiple anatomic structures across various planes, regions, and modalities.
• Current methods often require complex, multi-step processes for accurate segmentation.

Purpose of the Study:

• To introduce a novel holistic multitask regression network (HMR-Net) for improved medical image segmentation.
• To address the limitations of traditional clinical image analysis by formulating organ segmentation as a multitask learning problem.

Main Methods:

• HMR-Net employs multitask regression to estimate organ class, regional location, and precise contour coordinates simultaneously.
• The network utilizes hierarchical multiscale and fused organ features to capture nonlinear relationships and holistic shape information through coordinate correlations.

Main Results:

• HMR-Net achieved a mean average precision of 0.81 and a dice score of 0.93 on a representative multi-application database.
• The model demonstrated comparable or superior performance to state-of-the-art algorithms across eight datasets from 222 subjects.
• The method showed high-contrast sensitivity for delineating small and irregularly shaped organs.

Conclusions:

• HMR-Net provides an effective general framework for organ shape regression in multiple medical imaging applications.
• The model's flexible framework offers potential for fully automatic preliminary analysis of diverse medical images.
• The holistic approach enhances segmentation accuracy and efficiency.