A Federated Learning-Based CNN Architecture of Privacy-Preserving Osteoporosis Detection on Knee X-Ray Images

View at publisher Summarise with VIRAT
Authors: Shadnagar Dyaga Anushna, M.I. Thariq Hussan
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Knee X-ray images can be used to identify osteoporosis early and accurately to prevent the worsening of the bone and enhance patient outcomes. Conventional centralized deep learning methods are well-powered in their predictive capacity but demand the consolidation of a vast amount of medical data in one server- a method that is met with great privacy, security, and regulatory apprehensions. To overcome these constraints, this research paper presents a Federated Learning-based Convolutional Neural Network (FL-CNN) system that will be used to facilitate the joint training of models on multiple clients without necessarily sharing raw patient data. The data used in this study are the knee X-rays, which are applied to the samples to be used in training and testing. The FL-CNN model was compared to two control models: the Centralized CNN and the Local CNN. Although the Centralized CNN had the best accuracy (94.10 %), it is not suitable when privacy is at stake since the CNN will have all the data. An independently trained Local CNN at the individual client demonstrated lower performance$(88.52 \%)$, as the sample was not sufficiently diverse. The proposed FL-CNN, on the other hand, had 93.27 % accuracy, 92.10 % precision, 94.68 % recall, and 93.37 % F1-score, which is comparable to the performance of the centralized learning yet guarantees the utmost confidentiality of the data. The loss curve and the curriculum of accuracy also support the fact that the current model has the best convergence behavior and the ability to generalize, as opposed to the Local CNN. In general, the suggested FL-CNN solution is an efficient and privacy-friendly method of automated osteoporosis detection in distributed healthcare settings.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Medical Imaging and Analysis Bone health and osteoporosis research

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1733–1739
Publication date
21 January 2026
Language
EN