The goal of this observational study is to develop and validate a multimodal artificial intelligence prediction model for treatment-related complications in children with perimembranous ventricular septal defect (pmVSD) undergoing transcatheter device closure. The main question it aims to answer is: Can an AI model that integrates demographics, laboratory results, electronic health record text, echocardiography reports, chest radiographs, and electrocardiogram accurately predict the risk of complications at the individual patient level? Data will be retrospectively collected from routine clinical care records of pediatric patients who underwent transcatheter closure for pmVSD. Deep learning methods will be used to extract features from text and images to train and validate the prediction model.

Inclusion Criteria:

• Age ≤ 18 years at the time of transcatheter procedure.
• Diagnosis of perimembranous ventricular septal defect confirmed by echocardiography, and underwent transcatheter device closure at the study center.
• Medical records sufficient to ascertain the primary outcome within the pre-specified follow-up window, and availability of minimum baseline clinical information required for model development/validation.

Exclusion Criteria:

• Ventricular septal defects not classified as perimembranous on echocardiography, including muscular, outlet, or inlet VSDs, as well as multiple or complex VSDs involving more than one septal region.
• Presence of complex congenital heart disease or associated structural abnormalities requiring concomitant surgical repair (e.g., tetralogy of Fallot).
• Prior surgical VSD repair or prior transcatheter VSD closure.

Children (18 years or younger) with echocardiography-confirmed perimembranous ventricular septal defect who underwent transcatheter device closure at Xinhua Hospital between January 1, 2015, and December 31, 2025, identified retrospectively from routine clinical care records.

De-identified individual participant data will not be made publicly available due to privacy considerations and institutional data governance policies for retrospective electronic health record and imaging data. Aggregated results may be shared in publications and presentations.