Edge Computing for AI-Based Brain MRI Applications: A Critical Evaluation of Real-Time Classification and...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Edge Computing

Background:

• Medical imaging technologies (MRI, CT, PET, ultrasound) are crucial for diagnosis.
• Artificial Intelligence (AI) is increasingly used to enhance diagnostic radiology.
• Edge computing offers solutions for processing medical data in resource-limited remote areas.

Purpose of the Study:

• To evaluate the real-time deployment capabilities of various platforms for AI-driven diagnostic radiology tools.
• To compare the performance of different hardware and software configurations for medical image analysis applications.
• To conduct a cost-benefit analysis of deploying diagnostic tools on diverse computing platforms.

Main Methods:

• Utilized MRI classification and segmentation applications for performance testing.
• Evaluated platforms including a PC with NVIDIA GPU, Jetson Xavier NX, Raspberry Pi 4B, and Android phone.
• Used MATLAB, Python, and Android Studio for development and testing.
• Analyzed computational times and accuracy across different configurations.

Main Results:

• The Android phone achieved the fastest classification inference time (0.1068 s) using a TFLite model, with minimal accuracy loss.
• For segmentation, the Android phone (3.2023 s) and Jetson Xavier NX (5.2641 s) showed competitive performance.
• The Jetson Xavier NX and Android phone were identified as optimal platforms due to their balance of size, speed, and cost.

Conclusions:

• Edge computing platforms, particularly the Android phone and Jetson Xavier NX, are highly suitable for real-time AI diagnostic radiology applications.
• These platforms provide efficient and affordable solutions for medical image analysis, especially in remote or underserved regions.
• The study highlights the potential of accessible technology to advance diagnostic capabilities in healthcare.