Pediatric chest X-ray diagnosis using neuromorphic models.

Area of Science:

• Neuromorphic computing
• Artificial intelligence in medical imaging
• Biologically inspired computing

Background:

• Pediatric chest X-rays (PediCXR) are crucial for diagnosing thoracic illnesses in children.
• Current diagnostic methods can be time-consuming and require specialized expertise.
• There is a need for efficient and accurate automated diagnostic tools.

Purpose of the Study:

• To present an innovative neuromorphic method using Spiking Neural Networks (SNNs) for analyzing PediCXR.
• To evaluate the performance of SNN-based models in identifying prevalent pediatric thoracic illnesses.
• To demonstrate the potential of neuromorphic computing for real-time medical imaging diagnostics.

Main Methods:

• Utilized Spiking Convolutional Neural Networks (SCNN), Spiking Residual Networks (S-ResNet), and Hierarchical Spiking Neural Networks (HSNN).
• Employed spatiotemporal feature extraction, residual connections, and event-driven processing.
• Trained and evaluated models on the publically available benchmark PediCXR dataset.

Main Results:

• The Hierarchical Spiking Neural Network (HSNN) model achieved a classification accuracy of 96% across six thoracic illness categories.
• Achieved an F1-score of 0.95 and a specificity of 1.0 in pneumonia detection.
• HSNN model surpassed benchmark approaches from the literature in diagnostic precision.

Conclusions:

• Neuromorphic computing, using SNNs, is a feasible approach for pediatric chest X-ray analysis.
• Biologically inspired SNN models offer significant improvements in diagnostic accuracy and performance.
• This research paves the way for real-time, AI-powered medical imaging diagnostics.