OsteoNet—A Framework for Identifying Osteoporosis in Bone Radiograph Images Using Attention-Based VGG Network

OsteoNet—A Framework for Identifying Osteoporosis in Bone Radiograph Images Using Attention-Based VGG Network

Diagnosing osteoporosis from X-ray images poses a significant challenge due to the visual similarities between images from healthy subjects and patients. In this paper, we present a novel method for detecting osteoporosis. Our approach utilizes local phase quantization (LPQ) to identify fine texture...

Full description

Saved in:
Bibliographic Details
Main Authors: Abdul Wahab Muzaffar, Farhan Riaz, Muhammad Tahir
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Osteoporosis
convolutional neural networks
bone radiograph image
Online Access:https://ieeexplore.ieee.org/document/10872902/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Diagnosing osteoporosis from X-ray images poses a significant challenge due to the visual similarities between images from healthy subjects and patients. In this paper, we present a novel method for detecting osteoporosis. Our approach utilizes local phase quantization (LPQ) to identify fine texture variations that indicate changes in bone density, which are characteristic of osteoporosis. Furthermore, we augment the VGG network with an attention mechanism that incorporates both channel and spatial attentions. Channel attention is applied by calculating the mean and variance across channels, followed by a softmax operation to highlight essential features in the channel. Spatial attention is achieved through average pooling followed by exponential weighting to focus on relevant spatial regions within the images. A comparative analysis of the proposed method on ISBI 2014 challenge dataset against existing methods revealed that hand-crafted features mimicking gradients outperform those based on statistics, and CNN-based methods generally exhibit better performance. Our proposed method achieved an accuracy of 74.1%. These findings elucidate on the efficacy of integrating differential image features with attention mechanisms to improve the accuracy of osteoporosis identification from X-ray images.
ISSN:2169-3536

Similar Items