SPEMix: a lightweight method via superclass pseudo-label and efficient mixup for echocardiogram view classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Echocardiogram view classification is crucial for diagnosing heart diseases.
• Supervised methods struggle with generalization due to labeling difficulties.
• Semi-supervised methods face challenges with out-of-distribution data and complex models.

Purpose of the Study:

• To propose a novel open-set semi-supervised method (SPEMix) for echocardiogram view classification.
• To improve classification performance and generalization by leveraging out-of-distribution unlabeled data.
• To enable efficient clinical application with lightweight models.

Main Methods:

• Developed SPEMix with two core blocks: DAMix Block and SP Block.
• DAMix Block generates high-quality augmented echocardiograms using pixel-level masks.
• SP Block utilizes superclass probability distribution for pseudo-labeling unlabeled data.

Main Results:

• SPEMix improves classification accuracy by effectively utilizing unlabeled data.
• The method enhances generalization through superclass pseudo-labeling.
• A lightweight model trained with SPEMix achieved top performance on the TMED2 dataset.

Conclusions:

• SPEMix offers a robust solution for echocardiogram view classification, addressing limitations of current methods.
• The application of lightweight models in this domain facilitates clinical adoption.
• This approach aids cardiologists in more efficient and accurate heart disease diagnosis.