Automated assessment of cardiac pathologies on cardiac MRI using T1-mapping and late gadolinium phase sensitive inversion recovery sequences with deep learning

Abstract

Background: A deep learning (DL) model that automatically detects cardiac pathologies on cardiac MRI may help streamline the diagnostic workflow. To develop a DL model to detect cardiac pathologies on cardiac MRI T1-mapping and late gadolinium phase sensitive inversion recovery (PSIR) sequences were used.

Methods: Subjects in this study were either diagnosed with cardiac pathology (n = 137) including acute and chronic myocardial infarction, myocarditis, dilated cardiomyopathy, and hypertrophic cardiomyopathy or classified as normal (n = 63). Cardiac MR imaging included T1-mapping and PSIR sequences. Subjects were split 65/15/20% for training, validation, and hold-out testing. The DL models were based on an ImageNet pretrained DenseNet-161 and implemented using PyTorch and fastai. Data augmentation with random rotation and mixup was applied. Categorical cross entropy was used as the loss function with a cyclic learning rate (1e-3). DL models for both sequences were developed separately using similar training parameters. The final model was chosen based on its performance on the validation set. Gradient-weighted class activation maps (Grad-CAMs) visualized the decision-making process of the DL model.

Results: The DL model achieved a sensitivity, specificity, and accuracy of 100%, 38%, and 88% on PSIR images and 78%, 54%, and 70% on T1-mapping images. Grad-CAMs demonstrated that the DL model focused its attention on myocardium and cardiac pathology when evaluating MR images.

Conclusions: The developed DL models were able to reliably detect cardiac pathologies on cardiac MR images. The diagnostic performance of T1 mapping alone is particularly of note since it does not require a contrast agent and can be acquired quickly.

Keywords: Cardiomyopathies; Classification; Deep learning; Magnetic resonance imaging.

Fig. 1

Framework of the confusion matrix for a binary classification

Fig. 2

Performance of the model on the PSIR test set for classification as normal or abnormal. A shows the ROC curve of the model with an AUC value of 0.75. B illustrates the corresponding confusion matrix of the model with an overall accuracy of 88%

Fig. 3

Performance of the model on the T1-mapping test set for classification as normal or abnormal. A shows the ROC curve of the model with an AUC value of 0.69. B illustrates the corresponding confusion matrix of the model with an overall accuracy of 70%

Fig. 4

Heatmaps for cardiac pathology assessment on PSIR images. A, B: Subject without cardiac pathology. A shows the late gadolinium phase sensitive inversion recovery (PSIR) image. B shows a heatmap generated by overlaying a gradient-weighted class activation map (Grad-CAM) with the PSIR image. Red indicates higher activation, and blue indicates lower activation. The heatmap shows that the model mainly focused on the myocardial septum for its decision. This was classified by the deep learning model as normal with 86% certainty. C, D: Subject with chronic myocardial infarction. C shows the late gadolinium phase sensitive inversion recovery (PSIR) image. D shows a heatmap generated by overlaying a gradient-weighted class activation map (Grad-CAM) with the PSIR image. Red indicates higher activation, and blue indicates lower activation. The heatmap shows that the model mainly focused on the myocardium of the left ventricle, exhibiting wall thinning and an increase in signal intensity. The deep learning model diagnosed a cardiac pathology with 99% certainty

Fig. 5

Heatmaps for cardiac pathology assessment on T1 mapping images. A, B: Subject without cardiac pathology in the sagittal plane with short axis view. A shows the T1 mapping image. In B, the image was overlaid with a gradient-weighted class activation map (Grad-CAM), generating a heatmap. The heatmap depicts the focus areas of the model. Red indicates higher activation, and blue indicates lower activation. While making the classification, the network focused on parts of the image other than the heart. Thoracic muscles, spleen, intestines, and lower pole of the kidney represented the focus points. The model classified this case incorrectly as abnormal with 94% certainty. C, D: Subject with cardiac disease in the sagittal plane with short axis view. C shows the T1 mapping image. In D, the image was overlaid with a gradient-weighted class activation map (Grad-CAM), generating a heatmap. The heatmap depicts the focus areas of the model. Red indicates higher activation, and blue indicates lower activation. The strongest focus of the model was the right ventricle, including part of the septum. The kidney and liver represent weaker focus areas of the deep learning model. The network diagnosed a cardiac pathology with 100% certainty