Enhancing reginal wall abnormality detection accuracy: Integrating machine learning, optical flow algorithms, and...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Regional Wall Motion Abnormality (RWMA) is a critical early indicator of myocardial infarction (MI), a leading cause of mortality.
• Current automated echocardiography methods often analyze single-cycle data, potentially missing valuable information from multi-cycle and multi-view analyses.
• Enhanced RWMA detection is crucial for timely and effective MI treatment.

Purpose of the Study:

• To develop an innovative approach for RWMA detection by leveraging motion information across multiple echocardiographic cycles and views.
• To improve the accuracy and comprehensiveness of automated echocardiographic analysis for RWMA identification.

Main Methods:

• A three-phase algorithm utilizing U-Net for segmentation and optical flow for cardiac wall motion field features.
• Temporal Convolutional Networks (ConvNet), inspired by Temporal Segment Network (TSN), to interpret motion features.
• Machine learning and deep learning classifiers, including SVM, applied to A2C and A4C echocardiogram views.

Main Results:

• Support Vector Machine (SVM) classifier achieved high performance: 93.13% sensitivity, 83.61% specificity, 88.52% precision, and 90.39% F1 score.
• The method demonstrated an overall accuracy of 89.25% and an Area Under the Curve (AUC) of 95%.
• Results outperformed other studies using HMC-QU datasets, highlighting the method's effectiveness.

Conclusions:

• The developed technique offers a novel and precise tool for the early diagnosis of myocardial infarction.
• This approach enhances the capability of echocardiographic analysis for detecting RWMA more comprehensively.