Semi-Supervised Disease Classification Based on Limited Medical Image Data.

Area of Science:

• Medical Image Analysis
• Machine Learning
• Computer-Aided Diagnosis

Background:

• Semi-supervised learning with positive and unlabeled (PU) data is challenging in medical imaging due to limited labeled data.
• Existing methods struggle with the scarcity of annotated medical images for disease classification.
• PU learning for medical image-assisted diagnosis is crucial for reducing expert workload.

Purpose of the Study:

• To develop a novel generative model for semi-supervised disease classification using PU medical image data.
• To address the challenges posed by limited labeled medical images in classification tasks.
• To improve the accuracy and efficiency of medical image-aided diagnosis.

Main Methods:

• Introduction of a novel generative model inspired by Hölder divergence for PU learning.
• Comprehensive problem formulation and theoretical feasibility analysis.
• Extensive experiments on five benchmark medical image datasets (BreastMNIST, PneumoniaMNIST, BloodMNIST, OCTMNIST, AMD).

Main Results:

• The proposed Hölder divergence-based model significantly outperforms existing KL divergence-based methods.
• The novel approach achieves state-of-the-art performance across all five tested disease classification benchmarks.
• Demonstrated superiority in harnessing unlabeled medical images for improved classification.

Conclusions:

• The novel generative model offers a promising solution for semi-supervised disease classification with limited labeled medical data.
• The Hölder divergence approach effectively leverages unlabeled data, enhancing medical image analysis.
• This method represents a significant advancement in the field of medical image-aided diagnosis.