POD: PCM-Based Computing Platform for Object Detection in Biomedical Imaging Application

Recent advancements in image classification, object detection, and semantic segmentation have significantly acceler-ated medical image processing systems. However, for medical applications, these systems require exceptional performance to ensure reliability. High performance often entails increased computational load, power consumption, and system require-ments. In this paper, we investigate the potential of an efficient implementation of very deep object detection neural networks on Phase Change Memory (PCM)-based memristive crossbar circuits for MRI image labeling, aiming to reduce inference time and power consumption. The neural network model is developed from scratch, incorporating the latest breakthroughs in object detection, and is able to achieve up to 84.2%mAP performance on a standard brain tumor dataset containing 3064 T1-weighted contrast-inhanced images from 233 patients with meningioma, glioma, and pituitary tumor. We also discuss the design decisions, challenges, model performance on different model configurations, and issues that need to be addressed in future work.