KNEE OSTEOARTHRITIS STAGE CLASSIFICATION BASED ON HYBRID FUSION DEEP LEARNING FRAMEWORK
Knee osteoarthritis severity detection is one of the most challenging applications in computer vision due to the similarity between X-ray images of the adjacent stages. Handling huge number of X-ray images and the ability to detect the correct disease stage is based on advanced artificial intellige...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
University of Zakho
2025-04-01
|
| Series: | Science Journal of University of Zakho |
| Subjects: | |
| Online Access: | https://sjuoz.uoz.edu.krd/index.php/sjuoz/article/view/1450 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Knee osteoarthritis severity detection is one of the most challenging applications in computer vision due to the similarity between X-ray images of the adjacent stages. Handling huge number of X-ray images and the ability to detect the correct disease stage is based on advanced artificial intelligence technologies, like machine learning and deep learning. This study presents a novel deep learning-based fusion framework designed for detecting the severity of knee osteoarthritis and classifying its stages. The study utilizes two X-ray image datasets containing three challenges: imbalanced data, low contrast, and low data size. Data augmentation, adaptive histogram equalization, and limited oversampling techniques were used to solve these problems. Five deep learning architectures were utilized as base models (EfficientNetB0, EfficientNetV2B0, XceptionNet, ResNetRS101, and RegNetY032), followed by average pooling and dense layers. The feature-level, decision-level, score-level, and meta-based fusion technologies were also performed on the outputs of the best three trained models to minimize the individual models’ errors. The study registered 70% and 90.61% classification accuracies using both datasets. The study also found that the best models are the score-level and meta-based fusion models in all scenarios.
|
|---|---|
| ISSN: | 2663-628X 2663-6298 |