MV-MS-FETE: Multi-view multi-scale feature extractor and transformer encoder for stenosis recognition in...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Aortic stenosis is a prevalent heart valve disease, particularly in older adults.
• Early detection is critical to prevent irreversible progression and mortality.
• Echocardiogram analysis for stenosis can be hindered by imaging variations, leading to misdiagnosis.

Purpose of the Study:

• To develop an advanced deep learning model for improved aortic stenosis detection.
• To explore the benefits of multi-view echocardiogram analysis over single-view methods.
• To establish benchmarks for evaluating aortic stenosis recognition models.

Main Methods:

• A novel multi-view, multi-scale feature extractor and transformer encoder (MV-MS-FETE) architecture was proposed.
• The model was trained and tested on the Tufts medical echocardiogram public dataset.
• Performance was benchmarked against state-of-the-art single-view and multi-view models.

Main Results:

• The proposed MV-MS-FETE model demonstrated superior accuracy and F1-score compared to other multi-view methods.
• Multi-view approaches generally outperformed single-view methods in stenosis detection.
• The model exhibited stable performance during the training process.

Conclusions:

• The novel multi-view and multi-scale model effectively aids clinicians in diagnosing aortic stenosis.
• The study provides valuable benchmarks for future research in aortic stenosis recognition.
• Multi-view analysis is crucial for enhancing the accuracy of automated diagnostic systems.