Researchers have developed a deep learning framework named Ex-CMRNet, designed to automate the detection of myocardial inflammation from cardiac magnetic resonance (CMR) images. The study, published in BMC Medical Informatics and Decision Making, addresses the clinical challenge of interpreting CMR scans, a process that is often time-consuming and susceptible to inter-observer variability. By combining advanced neural network architectures with explainability tools, the team aimed to create a system that not only identifies abnormalities but also provides visual interpretations of its predictions.
The proposed framework utilizes a hybrid approach to extract visual features from medical images. It combines two distinct encoder models, EfficientNetB2 and InceptionV3, to capture complementary aspects of the visual data. These encoders work together to generate a rich set of representations that are then processed by a Multi-Head Self-Attention mechanism. This attention mechanism models informative relationships among the extracted representations.
To improve model robustness, training-time Gaussian noise regularization was incorporated. The best-performing configuration used a standard deviation of σ_train = 0.10.
A key component of the Ex-CMRNet framework is its integration of Local Interpretable Model-Agnostic Explanations, or LIME. By generating LIME explanations, the system provides visual interpretations of why it made a specific prediction.
The performance of Ex-CMRNet was evaluated using the Z-Alizadeh Sani dataset, a public collection of CMR images. The experimental results demonstrated that the model achieved high diagnostic accuracy. Specifically, the framework recorded an accuracy of 93.48% and a recall of 93.28%. The precision of the model was measured at 96.68%. Furthermore, the area under the receiver operating characteristic curve was 97.90%.
These metrics suggest that Ex-CMRNet provides a favorable balance between diagnostic performance, robustness, and interpretability.
Despite the promising results, the authors emphasize that the findings should be interpreted with caution due to specific limitations. The study relied on a single public dataset, which may limit the generalizability of the model to other populations or imaging protocols. Additionally, the absence of external, multicenter validation means that the model’s performance has not been tested across different clinical environments or with diverse patient demographics.
The study concludes that Ex-CMRNet represents a promising computer-assisted approach for the detection of myocardial abnormalities. The authors position the framework as a supportive tool that may assist, rather than replace, expert clinical assessment.
Subject of Research: Development of an explainable deep learning framework for automated myocardial inflammation detection in cardiac magnetic resonance imaging.
Article Title: CNN-based multi-head attention with explainability via LIME for myocardial diagnosis in cardiac magnetic resonance imaging
Article References: Shekarzaban, N., & Barzegar, B. (2026). CNN-based multi-head attention with explainability via LIME for myocardial diagnosis in cardiac magnetic resonance imaging. BMC Medical Informatics and Decision Making. https://doi.org/10.1186/s12911-026-03909-z
Image Credits: AI Generated
DOI: 10.1186/s12911-026-03909-z
Keywords: cardiac magnetic resonance, deep learning, myocardial inflammation, explainable AI, LIME, medical imaging, EfficientNetB2, InceptionV3, multi-head attention, diagnostic accuracy, CNN-based, multi-head
News Source: Ophelia Keating. (November 10, 2026). Explainable AI Model Detects Myocardial Inflammation in Cardiac MRI. Scienmag.



