Key result
Deep neural networks trained on a limited amount of data achieved high diagnostic performance, classifying myocardial delayed enhancement patterns on MR images with up to 82.1% accuracy.
Why the study?
To evaluate whether deep neural networks trained on a similar number of images required during physician training can acquire the capability to detect and classify myocardial delayed enhancement patterns.
Can deep neural networks accurately detect and classify myocardial delayed enhancement patterns on MR images?
Population
1995 MDE images from 200 consecutive patients undergoing cardiovascular MRI
Comparison
Three CNN architectures (GoogLeNet, AlexNet, ResNet-152) vs experienced cardiac MR image readers
Design
Retrospective feasibility study with fourfold cross-validation
Authors
Loading...
DNNs feasibly detect MDE patterns with limited images; leaves open prospective validation versus expert readers.
Observational (n=200)
Readers blinded to clinical information
No
Can deep neural networks accurately detect and classify myocardial delayed enhancement patterns on MR images?
Deep learning with convolutional neural networks can achieve high diagnostic performance in detecting and classifying myocardial delayed enhancement on MR images using a limited amount of training data.
Ohta et al. (2019) conducted an observational in Myocardial delayed enhancement (MDE) on cardiovascular MRI (n=200). Deep neural networks (CNNs) vs. Expert human readers (reference standard) was evaluated on Classification accuracy of MDE patterns (ResNet-152). Deep neural networks trained on a limited amount of data achieved high diagnostic performance, classifying myocardial delayed enhancement patterns on MR images with up to 82.1% accuracy.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: