Canine Echocardiography Dataset for MMVD Classification (Frame-Level and Case-Level Versions)

This dataset contains anonymized echocardiographic video frames extracted from 332 canine ultrasound examinations acquired between 2016 and 2024. Each case is assigned to one of three clinically verified diagnostic categories: Normal Heart Myxomatous Mitral Valve Disease (MMVD) without ruptured chordae tendineae (RCT) Myxomatous Mitral Valve Disease (MMVD) with RCT All ultrasound videos were exported from DICOM format, converted into individual RGB PNG frames, and fully anonymized. Preprocessing included removal of patient identifiers, ultrasound measurement overlays, ECG traces, and other textual artifacts. The dataset is provided in two complementary versions, corresponding to different stages of model development: 1. Frame-Level Dataset (CNN) Contains all frames stored individually and grouped by class.Designed for training and evaluating image-based Convolutional Neural Networks (CNNs) and for sequence-level prediction via max-voting. 2. Case-Level Dataset (CNN–LSTM) Contains exactly 30 preprocessed frames per case, stored in separate folders per dog.Designed for training temporal models, such as hybrid CNN–LSTM architectures. Metadata A metadata.csv file is included, containing: Case ID (anonymized) Diagnostic class Age Sex Breed No personal identifiers or sensitive information are included. Usage This dataset was used to train and evaluate both a ResNet50 CNN and a hybrid ResNet50–LSTM network for binary and multiclass classification of canine MMVD using echocardiography.