A Lightweight Fourier Block Transformer for Android-Based Edge-Enabled Detection of Osteopenia and Osteoporosis using X-ray Sensor Imaging Data

Abstract

The early detection of osteoporosis (OPRS) and osteopenia (OPNA) is crucial for preventing bone fractures and other bone-related complications in the aging population. The existing deep learning (DL) methods rely on cloud-based processing, which limits their suitability for point-of-care deployment for real-time detection of OPRS and OPNA using knee X-ray images. This letter proposes an Android-based edge-enabled lightweight Fourier block-based transformer (LFBBT) model for real-time detection of OPRS and OPNA diseases using knee X-ray images or X-ray sensor imaging data. The LFBBT model comprises a patch embedding layer, a discrete Fourier transform (DFT) block layer, a dense layer, a dropout layer, and an output layer. The knee X-ray images from a publicly available database are used to evaluate the performance of the proposed LFBBT model. The results show that the suggested LFBBT model has achieved an overall accuracy of 88.41%, which is higher than that of various transfer learning techniques and pre-trained transformers in detecting OPRS and OPNA diseases. The Android-based implementation of the LFBBT model has achieved a throughput of 60 images per minute for detecting OPRS and OPNA in real-time using knee X-ray images. © 2017 IEEE.