Interpretable machine learning for predicting isolated basal septal hypertrophy.

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Basal septal hypertrophy (BSH) is an under-recognized left ventricular morphological change with a 7-20% prevalence.
• BSH can indicate early left ventricular remodeling and poses risks for outflow tract obstruction and postoperative complications, especially after transcatheter aortic valve implantation (TAVI).
• Existing diagnostic and predictive models for BSH are limited, highlighting the need for advanced approaches like machine learning.

Purpose of the Study:

• To assess the efficacy of five machine learning algorithms in predicting basal septal hypertrophy (BSH).
• To develop a straightforward and effective prediction model for BSH using machine learning.

Main Methods:

• Echocardiographic and clinical data from 902 patients (91 BSH, 811 non-BSH) were analyzed.
• Five machine learning algorithms (XGBoost, Random Forest, Decision Tree, KNN, Naive Bayes) were applied, combined with Lasso regression-based logistic regression.
• Model performance was evaluated using ROC curves, calibration curves, and DCA; SHAP analysis interpreted model predictions.

Main Results:

• Logistic regression identified IVS-AO Angle, LVMI, LVIDdI, SBP, DBP, MVCP-Sd, GLU, and MV-A as associated with BSH.
• The XGBoost and Random Forest models achieved the highest AUCs in the test set (0.92 and 0.91, respectively).
• SHAP analysis revealed IVS-AO Angle, LVMI, LVIDdI, and SBP as key predictors, with IVS-AO Angle being the most significant.

Conclusions:

• Machine learning models can effectively predict basal septal hypertrophy (BSH).
• The IVS-AO Angle is identified as an independent predictor of BSH.
• The Random Forest (RF) model provides a simple and operable tool for BSH risk management.