MTW-ICHNet: Multi-task Weakly Supervised Learning with Enhanced Feature Descriptor Learning for Intracranial...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neurology

Background:

• Intracerebral hemorrhage (ICH) detection is crucial for patient outcomes.
• Weakly supervised learning (WSL) reduces reliance on extensive labeled data for ICH detection.
• Existing WSL methods struggle with efficient feature utilization and multi-task coordination.

Purpose of the Study:

• To develop an advanced multi-task WSL network, MTW-ICHNet, for improved intracerebral hemorrhage detection.
• To enhance feature discriminative capability and optimize collaborative learning across interdependent tasks.
• To address limitations in current WSL approaches for medical image analysis.

Main Methods:

• Introduced MTW-ICHNet, a multi-task WSL network integrating a feature descriptor and collaborative task optimization.
• Implemented feature enhancement techniques to refine extracted features during WSL training.
• Jointly optimized lesion localization and category recognition tasks within a unified architecture for cross-task knowledge sharing.

Main Results:

• Achieved 98.7% accuracy for hemorrhage classification and 97.5% for lesion localization.
• Demonstrated enhanced ICH image recognition through effective feature refinement and task collaboration.
• Validated improved performance in diagnostic accuracy under WSL conditions.

Conclusions:

• MTW-ICHNet effectively improves feature utilization and task collaboration in WSL for ICH detection.
• The proposed method offers accurate diagnostic references for patient-specific treatment strategies.
• Shows significant potential for clinical applications, especially in resource-limited settings with scarce annotations.