Optimizing Medical Image Classification Models for Edge Devices

Abstract

Machine learning algorithms for medical diagnostics often require resourceintensive environments to run, such as expensive cloud servers or high-end GPUs, making these models impractical for use in the field. We investigate the use of model quantization and GPU-acceleration for chest X-ray classification on edge devices. We employ 3 types of quantization (dynamic range, float-16, and full int8) which we tested on models trained on the Chest-XRay14 Dataset. We achieved a 2-4x reduction in model size, offset by small decreases in the mean AUC-ROC score of 0.0%-0.9%. On ARM architectures, integer quantization was shown to improve inference latency by up to 57%. However, we also observe significant increases in latency on x86 processors. GPU acceleration also improved inference latency, but this was outweighed by kernel launch overhead. We show that optimization of diagnostic models has the potential to expand their utility to day-to-day devices used by patients and healthcare workers; however, these improvements are context-and architecture-dependent and should be tested on the relevant devices before deployment in low-resource environments.

Local edge computing for radiological image reconstruction and computer-assisted detection: A feasibility study

“…Previous work exploring edge computing for breast cancer assessment using mammography image data could not be found. However, there are few papers describing the study of enabling technologies such as model optimization and hardware acceleration [10] and a blockchain-enabled learning model [11] with application examples in the medical imaging domain. Algorithms with low computational cost for medical image analysis have also been proposed [12].…”

“…The traditionally utilized, non-iterative Feldkamp-Davis-Kress (FDK) algorithm was used for reconstruction. The reconstructed slices were 1012-by-612 pixels in size, and the number of slices was modified (10,50,100,200) to address the influence of increasing reconstruction volume on the total computation time. Thus, the reconstructed volumes were 1012-by-612-by-…”

Lung Cancer Detection Model Using Deep Learning Technique

“…The data loading and preprocessing pipeline is optimized to reduce the training time. The author introduces quantization-aware training to maintain the trade-off between model size and inference speed [43]. Quantization-aware training optimizes the proposed model using weight and activation redundancy [43].…”

“…The author introduces quantization-aware training to maintain the trade-off between model size and inference speed [43]. Quantization-aware training optimizes the proposed model using weight and activation redundancy [43]. It allows the model to be lowered to a quarter of their original size and memory footprint.…”

“…It allows the model to be lowered to a quarter of their original size and memory footprint. In addition, the inference rates are increased by a factor of two to four with minimal or no loss of accuracy [43]. The training process enables the proposed model accessible for edge devices and real-time applications.…”