Abstract:We propose a novel anomaly detection method for echocardiogram videos. The introduced method takes advantage of the periodic nature of the heart cycle to learn three variants of a variational latent trajectory model (TVAE). While the first two variants (TVAE-C and TVAE-R) model strict periodic movements of the heart, the third (TVAE-S) is more general and allows shifts in the spatial representation throughout the video. All models are trained on the healthy samples of a novel in-house dataset of infant echocardiogram videos consisting of multiple chamber views to learn a normative prior of the healthy population. During inference, maximum a posteriori (MAP) based anomaly detection is performed to detect out-of-distribution samples in our dataset. The proposed method reliably identifies severe congenital heart defects, such as Ebstein's Anomaly or Shone-complex. Moreover, it achieves superior performance over MAP-based anomaly detection with standard variational autoencoders when detecting pulmonary hypertension and right ventricular dilation. Finally, we demonstrate that the proposed method enables interpretable explanations of its output through heatmaps highlighting the regions corresponding to anomalous heart structures.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of abnormal detection of congenital heart defects (CHDs) in neonatal and infant echocardiogram videos. Specifically, the researchers proposed a new variational autoencoder (VAE) - based method, called dynamic variational trajectory models (TVAE), to identify abnormalities in echocardiogram videos. #### Main problems: 1. **Data scarcity and imbalance**: Many CHDs, such as Ebstein’s Anomaly and Shone - complex, are very rare, making it impractical to collect large - scale labeled datasets, thus making supervised learning methods difficult to apply. 2. **Noise and complexity**: Echocardiogram video data is inherently noisy, and different types of CHDs have different manifestations, increasing the difficulty of detection. 3. **Automation requirements**: Traditional manual inspections are time - consuming and error - prone, and an automated abnormal detection method is required to improve efficiency and accuracy. #### Solutions: - **Train the model using healthy samples**: By training the model only with healthy echocardiogram videos, the need for large - scale datasets of rare diseases such as CHDs is avoided. - **Introduce periodic variational trajectory models (TVAE)**: Considering the periodicity of heart motion, the researchers proposed three variants of TVAE (TVAE - C, TVAE - R, and TVAE - S) to model the periodic and non - periodic motions of the heart. - **Maximum a posteriori probability (MAP) abnormal detection**: In the inference stage, MAP estimation is used for abnormal detection, which can effectively identify out - of - distribution samples, i.e., potential CHDs. - **Explanatory output**: Highlight the abnormal structure areas through heat maps to make the model output more explanatory and help doctors understand the detection results. #### Formula representation: Abnormality is defined as samples from low - probability regions: \[ A=\{x \in X; p_H(x) \leq \tau\} \] where \( p_H(x) \) is the probability density function of healthy samples, and \( \tau \) is the abnormal threshold. Reconstruction quality metric formula: \[ \alpha_f(x)=\| x - \hat{x} \|_2^2 \] where \( \alpha_f(x) \) represents the abnormal score of sample \( x \), and \( \hat{x} \) is the reconstruction result of sample \( x \) by model \( f \). In conclusion, this paper proposes an innovative deep - learning - based abnormal detection method specifically for echocardiogram videos, which can effectively identify multiple CHDs in the absence of large - scale labeled data and provide interpretable detection results.

Anomaly Detection in Echocardiograms with Dynamic Variational Trajectory Models

Human Kinematics-inspired Skeleton-based Video Anomaly Detection

Approaches Toward Physical and General Video Anomaly Detection

Video Anomaly Detection via Spatio-Temporal Pseudo-Anomaly Generation : A Unified Approach

Hallucinated Heartbeats: Anomaly-Aware Remote Pulse Estimation

Video anomaly detection based on a multi-layer reconstruction autoencoder with a variance attention strategy

Video Trajectory Classification and Anomaly Detection Using Hybrid CNN-VAE

Unsupervised Detection of Anomalies in Fetal Heart Rate Tracings using Phase Space Reconstruction

Anatomy-Aware Cardiac Motion Estimation

Multimodal Variational Autoencoder for Low-cost Cardiac Hemodynamics Instability Detection

A Novel Temporal Generative Adversarial Network for Electrocardiography Anomaly Detection

Learning Attention Augmented Spatial-temporal Normality for Video Anomaly Detection

Automated Recognition of Regional Wall Motion Abnormalities Through Deep Neural Network Interpretation of Transthoracic Echocardiography

Anomaly Detection in Optical Coherence Tomography Angiography (OCTA) with a Vector-Quantized Variational Auto-Encoder (VQ-VAE)

Deep reinforcement learning and convolutional autoencoders for anomaly detection of congenital inner ear malformations in clinical CT images

EB-GAME: A Game-Changer in ECG Heartbeat Anomaly Detection

An Exploratory Study on Human-Centric Video Anomaly Detection through Variational Autoencoders and Trajectory Prediction

Learning normal appearance for fetal anomaly screening: Application to the unsupervised detection of Hypoplastic Left Heart Syndrome

TeVAE: A Variational Autoencoder Approach for Discrete Online Anomaly Detection in Variable-state Multivariate Time-series Data

Finding "Good Views" of Electrocardiogram Signals for Inferring Abnormalities in Cardiac Condition

Anomalies cannot materialize or vanish out of thin air: A hierarchical multiple instance learning with position-scale awareness for video anomaly detection