Related Experiment Video
Updated: Nov 1, 2025
12:10
Semi-automated Optical Heartbeat Analysis of Small Hearts
Published on: September 16, 2009
12.5K
Deep Learning Algorithm to Detect Cardiac Sarcoidosis From Echocardiographic Movies.
Susumu Katsushika1, Satoshi Kodera1, Mitsuhiko Nakamoto1
1Department of Cardiovascular Medicine, The University of Tokyo Hospital.
Summary
A deep learning algorithm using echocardiographic movies can effectively distinguish cardiac sarcoidosis (CS) patients from healthy individuals. This artificial intelligence tool shows promise for early CS detection, matching cardiologist performance.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Imaging
Background:
- Early diagnosis of subclinical cardiac sarcoidosis (CS) is challenging.
- Echocardiography is a key imaging modality for cardiac assessment.
Purpose of the Study:
- To develop and evaluate a deep learning algorithm for distinguishing CS patients from healthy subjects using echocardiographic movies.
- To assess the performance of a transfer learning-based 3D convolutional neural network (3D-CNN) for CS detection.
Main Methods:
- Two 3D-CNN models (pretrained and non-pretrained transfer learning) were trained on 212 echocardiographic movies (50 CS patients, 149 healthy subjects).
- Model performance was evaluated on an independent set of 41 echocardiographic movies.
- Algorithm performance was compared to the interpretations of 5 cardiologists.
Main Results:
- The pretrained 3D-CNN algorithm achieved a higher area under the receiver-operating characteristic curve (AUC) (0.842) compared to the non-pretrained algorithm (0.724).
- The pretrained algorithm's AUC was not significantly different from that of cardiologists (0.855).
- Sensitivity analysis indicated the algorithm focused on the mitral valve region.
Conclusions:
- A 3D-CNN utilizing transfer learning is a promising tool for detecting cardiac sarcoidosis from echocardiographic movies.
- This AI approach may aid in the early diagnosis of subclinical CS.
