Open Set Medical Diagnosis via Difficulty-Aware Multi-Label Thorax Disease Classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Emerging diseases and global outbreaks like COVID-19 highlight the need for advanced diagnostic tools.
• Medical imaging analysis presents unique challenges, often involving multi-label classification where multiple diseases can coexist.
• Existing open-set recognition (OSR) methods are unsuitable for medical diagnosis due to the inherent multi-label nature and the classification of non-exceeding thresholds as 'normal' rather than 'unknown'.

Purpose of the Study:

• To propose a novel method for open-set medical diagnosis tailored to the complexities of multi-label classification.
• To address the fundamental limitations of current OSR techniques in the medical domain.
• To improve the accuracy and reliability of automated medical diagnostic systems.

Main Methods:

• Development of a novel open-set medical diagnosis approach.
• Utilization of Copycat and entropy-based thresholds to handle multi-label classification challenges.
• Adaptation of OSR principles for the specific constraints of medical image analysis.

Main Results:

• The proposed method demonstrates strong performance in multi-label classification tasks.
• The approach effectively recognizes both normal and unknown conditions within medical imaging datasets.
• Experimental validation confirms the efficacy of the novel method in addressing open-set medical diagnosis.

Conclusions:

• The developed method offers a significant advancement for open-set multi-label medical diagnosis.
• This research pioneers a solution for a previously unaddressed problem in medical AI.
• The findings pave the way for more robust and accurate automated medical diagnostic systems.