International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 11 | Nov 2025
p-ISSN: 2395-0072
www.irjet.net
Fracture Diagnosis Approach Using Deep Learning Kamala R1, Nevitha Teja Choudera2, Pooja B N3, Anusha G K4, Karthik K5 1
Assistant Professor, Information Science and Engineering, Bapuji Institute of Engineering and Technology, Davangere, affiliated to VTU Belagavi, Karnataka, India. 2,3,4,5 Bachelor of Engineering, Information Science and Engineering, Bapuji Institute of Engineering and Technology, Davangere, affiliated to VTU Belagavi, Karnataka, India. ---------------------------------------------------------------------------------***----------------------------------------------------------------------------------Abstract - The Bone fractures are among the most frequently encountered orthopedic injuries and require rapid and accurate
diagnosis for effective treatment. Conventional radiographic interpretation depends on manual assessment by radiologists, which is time-consuming, subjective, and prone to diagnostic variability, especially in high-workload clinical environments. This paper proposes an automated bone fracture detection system using Convolutional Neural Networks (CNNs) to classify Xray images into fractured and non-fractured categories. The system integrates both a custom CNN architecture and transfer learning models, including ResNet-50, DenseNet-121, and EfficientNetB3, to enhance diagnostic accuracy. Pre-processing steps such as Contrast Limited Adaptive Histogram Equalization (CLAHE), Gaussian filtering, and data augmentation techniques are applied to improve feature extraction and reduce noise. The implementation is supported by a Flask-based clinical interface enabling secure image upload and real-time prediction for practical usability. The custom CNN model achieved an overall accuracy of 92.5%, demonstrating reliable baseline performance. The EfficientNetB3 transfer learning model outperformed other architectures with an accuracy of 96.2%, sensitivity of 98.7%, and specificity of 94.1% on the test dataset. These results indicate superior fracture localization capability and reduced false-negative rates, which are critical for clinical adoption. The study highlights that ensemble transfer learning significantly enhances prediction performance when compared to traditional machine learning and standalone CNN approaches. The proposed system reduces radiologist workload, accelerates diagnostic turnaround time, and supports computer-aided diagnosis (CAD) in orthopedic imaging. This research further demonstrates the potential for deployment in resource-limited settings where access to expert radiologists is scarce. Standardized fracture detection, automation of routine evaluation, and an accessible web- based tool contribute to improved healthcare delivery. Key Words: Bone Fracture Detection, Convolutional Neural Networks (CNN), Transfer Learning, Deep Learning, Medical Image Analysis, Computer-Aided Diagnosis (CAD)
1. INTRODUCTION Bone fractures occur in millions of people worldwide every year, and diagnosis often relies heavily on imaging (X-rays, CT scans, or MRIs). Historically, assessing fractures has relied on the expertise of radiologists. Due to heavy dependence on radiologists to assess images and reach a diagnosis, this practice can create a bottleneck in emergency departments or imaging centers. Evidence suggests radiologist error rates can range from 10-25% in detecting fractures (especially in subtle or complicated cases). The availability of radiologists is often limited in developing nations, leading to increased wait times for patients and increased morbidity and mortality. Artificial Intelligence (AI) and deep learning have improved the capability of medical professionals to analyze medical images, and convolutional neural networks (CNNs) have outperformed humans in many visual pattern detection tasks and have achieved comparable, or better, diagnostic accuracy in various other tasks. Recent literature(20232025) reports that CNNs can detect fractures with 95-99% accuracy across diverse datasets. CNNs benefit from transfer learning, which uses a CNN retrained on the Imagine dataset, dramatically reducing the amount of time and computational power needed to train the model. This project presents an automated Bone Fracture Detection System using a custom CNN and transfer learning (Res Net- 50, DenseNet-121, and EfficientNetB3) to provide binary classification (fractured/non-fractured) of bone fractures and provide performance metrics for clinical practice. One significant innovative aspect was the inclusion of a simple web application (Flask) to illustrate functional deployment of these data models so that a radiologist could easily use it in practice.
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 975