Browsing by Author "MUHAMMAD HASEEB BILAL"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    SEVERITY BASED CLASSIFICATION OF KNEE OSTEOARTHRITIS FROM X-RAY IMAGES USING DEEP LEARNING NOVEL FUSION APPROACH
    (UMT, Lahore, 2023) MUHAMMAD HASEEB BILAL
    Knee osteoarthritis is a most common degenerative sort of joint swelling and tenderness. It is characterized by degeneration of the knee's articular cartilage the adaptable, evasive material that regularly shields bones from joint grinding and effect. It happens most frequently in middle aged people, in spite of the fact that it might happen in more youthful individuals, as well. In essential medical care, knee osteoarthritis is analyzed utilizing clinical assessment and radiographic appraisal. This study aims to classify knee osteoarthritis based on severity levels via knee x-rays to assist orthopedic surgeons through using deep learning models. Classifiers have been made for differentiating between healthy to severe knee x-ray images. The dataset is consisted on 5 classes from grade 0 to 4 based on the intensity of injury. The available dataset is highly imbalanced. The dataset has been balanced using different augmentation techniques. This study employed a step-wise approach in developing knee osteoarthritis classification models. Initially, convolutional neural network (CNN) based classifiers have been used. Then, pre-trained deep learning models were employed with transfer learning techniques. Subsequently, transformer neural networks have been utilized and they outperformed in terms of evaluation matrices. Furthermore, a novel architecture has been proposed which utilized ConvNext and Swin Transformer Feature Extractor to classify X-ray images through softmax layer. All these models have been trained using both imbalance and balance datasets, resulting in gradual improvements in the accuracy of the results. The accuracies for deep transfer learning models with balanced dataset are 71%, 72%, 54%, 68%, 69%, 59% and 61% for models named MobileNet, ResNet-50, VGG19, InceptionV3, InceptionResNetV2, Xception and Efficient-Net respectively.