A deep learning-based single-view echocardiographic analysis for prediction of left ventricular outflow tract obstruction after transcatheter aortic valve implantation

Background/Introduction Transcatheter aortic valve implantation (TAVI) is widely performed in patients with severe degenerative aortic stenosis (AS). Although TAVI generally improves stroke volume by reducing left ventricular afterload, a subset of patients develop left ventricular outflow tract obstruction (LVOTO) after the procedure. Post-TAVI LVOTO is a rare but potentially life-threatening complication associated with acute haemodynamic instability. We previously developed a deep learning-derived index for predicting LVOTO in hypertrophic cardiomyopathy using only parasternal long-axis (PLAX) transthoracic echocardiography (TTE) images. Purpose To evaluate the clinical utility of a deep learning-based LVOTO index derived from pre-procedural TTE for predicting post-TAVI LVOTO. Methods We retrospectively analysed 301 consecutive patients who underwent TAVI at a tertiary referral centre between 2017 and 2025. A pre-trained deep learning model was applied to PLAX views extracted from pre-procedure TTE to capture morphological and dynamic motion features of the left ventricle, generating a deep learning index for LVOTO (DLi-LVOTO; range 0–100). Discriminative performance for post-TAVI LVOTO was evaluated using the area under the receiver operating characteristic curve (AUROC). Independent association with post-TAVI LVOTO was assessed using multivariable logistic regression adjusted for established clinical and echocardiographic risk factors. Results DLi-LVOTO was positively correlated with baseline LVOTO severity on pre-procedural TTE. Post-TAVI LVOTO occurred in 42 patients (14.0%). Patients who developed post-TAVI LVOTO had significantly higher DLi-LVOTO scores than those who did not (median 54 vs. 17; P<0.001). DLi-LVOTO demonstrated good discriminative performance for predicting any post-TAVI LVOTO (AUROC 0.782) and moderate-or-greater LVOTO (AUROC 0.736) (Figure 1). After adjustment for conventional risk factors, including small left ventricular cavity size and baseline LVOTO, DLi-LVOTO remained independently associated with post-TAVI LVOTO (adjusted OR 1.31 per 10-point increase; 95% CI 1.08–1.58; P=0.005). Consistent associations were observed in patients without baseline LVOTO on pre-procedure TTE. Conclusions A deep learning-derived index obtained from a single PLAX echocardiographic view is associated with post-TAVI LVOTO and demonstrates good discriminative performance. DLi-LVOTO may serve as a complementary tool for identifying patients at risk of post-TAVI LVOTO during pre-procedural evaluation.