International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 12 | Dec 2025
p-ISSN: 2395-0072
www.irjet.net
Bone Fracture Detection and Classification from 3D Medical Imaging Using 3D-CNN Parveen Taj S1, Abhishek S M2, Karthik S L3, Prajwalagouda Jakkangoudar4, Pruthviraj N P5 1 Assistant Professor, Information Science and Engineering, Bapuji Institute of Engineering and technology,
Karnataka, India 2,3,4,5, Bachelor of Engineering, Information Science and Engineering, Bapuji Institute of Engineering and
technology, Karnataka, India ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Bone fracture diagnosis from CT and MRI
volumetric inputs. While traditional machine learning approaches rely heavily on handcrafted features and semiautomatic procedures, 3D-CNNs automatically learn structural attributes of bone morphology, enabling precise classification and localization of fractures.
scans is often time-consuming and prone to human error due to the complexity of analyzing 3D anatomical structures. This paper presents a fully automated 3D Convolutional Neural Network (3D-CNN) capable of detecting and classifying fractures directly from volumetric medical images. The system performs standardized preprocessing, reconstructs 3D voxel inputs, and generates both fracture predictions and voxel-level localization heat maps. Experimental results show high classification accuracy and reliable localization, demonstrating that the proposed 3D-CNN framework offers an efficient and clinically viable alternative to manual diagnostic methods.
This paper presents a 3D-CNN powered fracture detection system that automates the entire diagnostic pipeline: preprocessing, segmentation, classification, and localization. The system mimics the clarity and diagnostic reasoning of radiologists while significantly reducing manual effort.
2. PROPOSED SYSTEM
Key Words: 3D-CNN, Bone Fracture Detection, Deep Learning, CT scan Analysis, Medical Imaging, Fracture Localization.
The proposed architecture introduces an end-to-end deep learning pipeline for processing volumetric medical images and predicting fracture type and location.
1. INTRODUCTION
Key Components
Bone fractures are among the most common orthopaedic injuries, affecting millions globally each year. Accurate and timely fracture diagnosis is essential for planning treatment and preventing further complications. Conventional diagnostic workflows primarily rely on 2D X-ray interpretation, a process that is prone to human fatigue, subjectivity, and inconsistency. While X-rays serve as the first-line diagnostic modality, they lack volumetric depth and often fail to reveal subtle or overlapping fractures—especially in regions such as wrists, ankles, and ribs.
1. 3D Medical Imaging Input Unit: CT or MRI scan volumes are imported in DICOM format and converted into standardized voxel grids. The module performs orientation correction, slice alignment, and Hounsfield-Unit (HU) normalization to ensure uniform 3D input quality across different scanners and hospitals.
Advancements in Computed Tomography (CT) and Magnetic Resonance Imaging (MRI) provide complete 3D anatomical information, enabling significantly improved diagnostic clarity. However, analyzing hundreds of CT/MRI slices manually is time-consuming and cognitively demanding for radiologists. This creates the need for automated, intelligent systems capable of processing volumetric medical data efficiently and accurately. Recent progress in deep learning—specifically 3D Convolutional Neural Networks—has shown remarkable capability in detecting spatial patterns across
© 2025, IRJET
|
Impact Factor value: 8.315
|
2.
3D Preprocessing & Bone Enhancement Module: The volumetric scan is refined using 3D Gaussian filtering, contrast enhancement, and bone-region extraction. These steps amplify subtle fracture lines and remove soft-tissue noise, ensuring that the deep-learning model receives a clean and consistent volumetric representation of the bone structures.
3.
3D-CNN Fracture Detection Model: A customized 3D Convolutional Neural Network processes the vowelized input to detect fractures automatically. The model learns spatial depth patterns across slices, enabling accurate identification of fracture types such as hairline, displaced, and comminuted fractures.
ISO 9001:2008 Certified Journal
|
Page 235