Su-MICL: Severity-Guided Multiple Instance Curriculum Learning for Histopathology Image Interpretable Classification.

Area of Science:

• Computational pathology
• Digital histopathology
• Machine learning in medicine

Background:

• Histopathology image classification is vital for clinical diagnosis but often lacks interpretability.
• Current methods requiring detailed lesion-level annotations are impractical.
• Multiple-instance learning (MIL) offers potential but has limitations in accuracy and prior information requirements.

Purpose of the Study:

• To develop an interpretable histopathology image classification strategy that avoids tedious lesion-level labeling.
• To leverage disease severity as a novel prior within a MIL framework.
• To enhance lesion identification accuracy in weakly supervised settings.

Main Methods:

• Proposed a novel severity-guided multiple instance curriculum learning (Su-MICL) strategy.
• Integrated disease severity to define image learning difficulty.
• Implemented a curriculum learning approach, training the model from easy to hard examples.

Main Results:

• Su-MICL achieved performance comparable to state-of-the-art weakly supervised methods for image-level classification.
• Lesion identification performance closely approached that of supervised learning methods.
• Demonstrated effectiveness on two histopathology image datasets.

Conclusions:

• The Su-MICL strategy provides an interpretable diagnosis without laborious lesion annotation.
• This approach offers valuable insights for histopathology image diagnosis.
• Severity-guided curriculum learning enhances the utility of MIL in computational pathology.