Related Experiment Video
Updated: Sep 11, 2025
Retrospective Cardiac Gating with A Prototype Small-Animal X-ray Computed Tomograph
Published on: February 21, 2025
BM3D filtering with Ensemble Hilbert-Huang Transform and spiking neural networks for cardiomegaly detection in chest
1School of Computing Science and Engineering, VIT Bhopal University, Sehore, India.
Insights
This study introduces an automated method for diagnosing cardiomegaly using X-ray images. A Spiking Neural Network (SNN) achieved high accuracy, improving early detection of heart conditions.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Cardiology
Background:
- Cardiomyopathy is a severe condition leading to heart failure and sudden cardiac death.
- Cardiomegaly, or heart enlargement, is an early indicator of cardiac disease detectable via chest X-ray.
- Timely diagnosis of cardiomegaly is crucial for effective treatment and improved patient prognosis.
Purpose of the Study:
- To develop an automated diagnostic methodology for cardiomegaly using X-ray images.
- To evaluate the performance of various machine learning models, including Spiking Neural Networks (SNNs), for this task.
Main Methods:
- A novel approach fusing Block-Matching and 3D Filtering (BM3D) with Ensemble Hilbert-Huang Transform (EHHT) for feature extraction from X-ray images.
- Implementation and comparison of deep learning models (VGG16, ResNet50, InceptionV3, DenseNet169) and Spiking Neural Networks (SNNs).
- Utilized Gradient-weighted Class Activation Mapping (GradCAM) to visualize model decision-making processes.
Main Results:
- The Spiking Neural Network (SNN) demonstrated superior performance compared to other models.
- The SNN achieved an accuracy of 97.6%, sensitivity of 96.3%, and specificity of 98.2% in diagnosing cardiomegaly.
- GradCAM highlighted key image regions influencing the SNN's diagnostic predictions.
Conclusions:
- The developed SNN-based methodology shows significant potential for accurate and efficient automated diagnosis of cardiomegaly from X-ray images.
- This approach can enhance clinical decision-making and improve patient outcomes in managing heart conditions.
- Further research into SNNs for medical image analysis is warranted.
Abstract:
Cardiomyopathy is a life-threatening condition associated with heart failure, arrhythmias, thromboembolism, and sudden cardiac death, posing a significant contribution to worldwide morbidity and mortality. Cardiomegaly, which is usually the initial radiologic sign, may reflect the progression of an underlying heart disease or an underlying undiagnosed cardiac condition. Chest radiography is the most frequently used imaging method for detecting heart enlargement. Prompt and accurate diagnosis is essential for prompt intervention and appropriate treatment planning to prevent disease progression and improve patient outcomes. The current work provides a new methodology for automated cardiomegaly diagnosis using X-ray images through the fusion of Block-Matching and 3D Filtering (BM3D) within the Ensemble Hilbert-Huang Transform (EHHT), convolutional neural networks like Pretrained VGG16, ResNet50, InceptionV3, DenseNet169, and Spiking Neural Networks (SNN), and Classifiers. BM3D is first used for image edge retention and noise reduction, and then EHHT is applied to obtain informative features from X-ray images. The features that have been extracted are then processed using an SNN that simulates neural processes at a biological level and offers a biologically possible classification solution. Gradient-weighted Class Activation Mapping (GradCAM) emphasized important areas that affected model predictions. The SNN performed the best among all the models tested, with 97.6 % accuracy, 96.3 % sensitivity, and 98.2 % specificity. These findings show the SNN's high potential for facilitating accurate and efficient cardiomyopathy diagnosis, leading to enhanced clinical decision-making and patient outcomes.
