SCMR-ISMRM Workshop: 05 - Scientific Abstracts (AI and Machine Learning & Workflow)-Lecture 4 - Deep learning for classification of cine CMR images: the first step for a fully automated CMR analysis

Lecture 4 - Deep learning for classification of cine CMR images: the first step for a fully automated CMR analysis

Video Transcription

Video Summary

Researchers at King’s College London developed a deep learning framework to automatically identify standard Cine CMR images (2-, 3-, 4-chamber, and short-axis) from routine cardiac MRI scans. Using over 100,000 images from 3,500 subjects across multiple vendors, the system first removed still images, then used a CNN to classify the rest. Trained on expert-labeled data, the best model achieved near-100% accuracy. The framework was later combined with quality control and cardiac function analysis, creating a rapid, fully automated pipeline that takes 4–7 minutes and detects poor-quality images with high sensitivity.

Keywords

deep learning

cine CMR

cardiac MRI

image classification

quality control