Automated LVH Grading: Integration of Deep Learning and Explainable AI for Accurate Diagnosis

Accurate grading of Left Ventricular Hypertrophy (LVH) is crucial for effective disease management. Echocardiography, surpassing ECG in sensitivity, is the preferred diagnostic tool for LVH grading, aiding in the detection of associated anomalies. Grading LVH requires expert interpretation; however, challenges in manual grading by echocardiographers introduce inconsistencies in clinical management and pose risks of misdiagnosis. In response to these challenges, this study presents a pioneering methodology that utilizes computer vision techniques for LVH grading, integrating information from echocardiography images and numerical patient data. Leveraging advanced deep learning and machine learning approaches, the objective is to enhance the accuracy and efficiency of LVH grading while mitigating complexities associated with manual measurements and interpretative subjectivity. The novel methodology employs transfer learning models (VGG16, ResNet, VGG19, GoogleNet, AlexNet) for feature extraction. Explainable AI models (LIME, SHAP) are incorporated to enhance interpretability, while the Synthetic Minority Over-sampling Technique (SMOTE) addresses class imbalance in merging features from images and numerical data. This comprehensive approach aims to standardize and automate LVH grading, ultimately improving transparency and reliability in clinical outcomes.