Osteoarthritis Detection Using Densely Connected Neural Network

Chaugule, Sushma V.; Malemath, V. S.

doi:10.1007/978-3-031-07005-1_9

Cited by 3 publications

(1 citation statement)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…With the emergence of ML and Deep Neural Network (DNN) models, various advanced techniques have been developed for classifying OA [25][26][27] , because of their ability to detect complex patterns of OA and provide accurate results. [28][29][30] used a densely connected convolutional neural network (DenseNet) to classify KOA severity from plain radiography. The proposed model results showed an accurate and precise diagnosis of OA.…”

Section: Related Workmentioning

confidence: 99%

Towards identifying key features in the classification of knee osteoarthritis – An enhanced feature fusion based deep network model

Chaugule¹,

Malemath²

2023

Preprint

View full text Add to dashboard Cite

Osteoarthritis is a gradual loss and destruction of articular cartilage, whose symptoms can be managed but not reversed. Early and accurate severity detection of Knee Osteoarthritis (KOA), from digital X-ray images is an essential need. A Deep Convolutional Neural Network (DCNN) hinged on feature extraction and fusion, to classify is presented to grade the severity according to Kellgren-Lawrence (KL) grading system. This work is taken through 4 stages; i) Autoencoder based de-noising of the X-ray images. ii) Segmentation to split the image into exhausted regions. iii) Feature extraction processes are employed to extract four types of features: region, Zernike, wavelet, and Haralick, which are further fused to enhance feature representation. iv) The outcome of the feature fusion is passed to a DCNN model that is used to cognize the knee area and classify it into appropriate severity grades. This model is trained using an Adam optimization algorithm, tested, validated experimentally concerning different evaluation metrics, and achieved testing and validation accuracies of 96.31% and 95.70% respectively. This model is based on feature extraction, enhancing feature representation, and identifying key features in the classification of KOA that outperforms all other models in detecting the severity of KOA.

show abstract