Comparative Analysis of Deep Convolutional Neural Network for Detection of Knee Injuries

Comparative Analysis of Deep Convolutional Neural Network for Detection of Knee Injuries
  IJETT-book-cover           
  
© 2024 by IJETT Journal
Volume-72 Issue-2
Year of Publication : 2024
Author : Anita Thengade, Archana Rajurkar
DOI : 10.14445/22315381/IJETT-V72I2P106

How to Cite?

Anita Thengade, Archana Rajurkar, "Comparative Analysis of Deep Convolutional Neural Network for Detection of Knee Injuries," International Journal of Engineering Trends and Technology, vol. 72, no. 2, pp. 47-57, 2024. Crossref, https://doi.org/10.14445/22315381/IJETT-V72I2P106

Abstract
Commonly occurring knee injuries such as Anterior Cruciate Ligament and Meniscus Torn lead to osteoarthritis problems in people. Radiologists very often recommend Magnetic Resonance Imaging for diagnosis of knee injuries. However, longer MRI interpretation time, vulnerability to clinical errors, and inconsistency are the major issues in the application of the MRI. The high volume of imaging and complexity of the patient’s profile make the task time-consuming, thereby increasing the workload of radiologists. Deep Learning-based automated techniques can help radiologists identify high-risk patients and aid as a support system for decision-making. In this study, we focus on different types of pre-trained networks to perform the classification of Knee Magnetic Resonance Images. The proposed work comprises three different classifiers, viz. knee abnormality, meniscus tear, and ACL tear, for independently classifying these three labels. In the proposed framework, features from knee Magnetic Resonance Images (MRI) are extracted using three well-known backbone networks, namely NASNet Large, NASNet Mobile, and ResNet50, for classification purposes. We also propose a Deep Convolutional Neural Network (DCNN) with residual block and NASNet Large as a feature extractor. The performance of these networks is evaluated for the MRNet dataset published by the Stanford ML Group. Overall, we achieved a performance of 94.47% for knee abnormalities for NASNet Large on the sagittal plane, and ResNet50 achieved ACL accuracy of 91.78% on the sagittal plane. For meniscus tear detection, the proposed DCNN model outperformed the state-of-the-art with a performance of 85.82% on the axial plane. We found that the proposed framework and carefully fine-tuned the network architecture were crucial factors in determining the best performance.

Keywords
Knee injuries, Deep Convolutional Neural Network, MRNet dataset.

References
[1] Alexander Selvikvag Lundervold, and Arvid Lundervold, “An Overview of Deep Learning in Medical Imaging Focusing on MRI,” Journal of Medical Physics, vol. 29, no. 2, pp. 102-127, 2019.
[CrossRef] [Google Scholar] [Publisher Link]
[2] Boehringer-Ingelheim, Radiology Rounds; Role of HRCT in Diagnosing Interstitial Lung Disease (ILD). [Online]. Available: https://www.ipfradiologyrounds.com/hrct-primer/image-reconstruction/
[3] R.J.M. Bruls, and R.M. Kwee, “Workload for Radiologists during On-Call Hours: Dramatic Increase in the Past 15 Years,” Insights into Imaging, vol. 11, no. 121, pp. 1-7, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[4] Kamel Rahouma, and Ahmed Salama, “Knee Images Classification Using Transfer Learning,” Procedia Computer Science, vol. 194, pp. 9-21, 2021.
[CrossRef] [Google Scholar] [Publisher Link]
[5] Nicholas Bien et al., “Deep-Learning-Assisted Diagnosis for Knee Magnetic Resonance Imaging: Development and Retrospective Validation of MRNet,” PLoS Medicine, vol. 15, no. 11, pp. 1-19, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[6] Ali Can Kara, and Firat Hardalaç, “Detection and Classification of Knee Injuries from MR Images Using the MRNet Dataset with Progressively Operating Deep Learning Methods,” Machine Learning and Knowledge Extraction, vol. 3, no. 4, pp. 1009-1029, 2021.
[CrossRef] [Google Scholar] [Publisher Link]
[7] David Azcona, Kevin McGuinness, and Alan F. Smeaton, “A Comparative Study of Existing and New Deep Learning Methods for Detecting Knee Injuries Using the MRNet Dataset,” 2020 International Conference on Intelligent Data Science Technologies and Applications (IDSTA), Valencia, Spain, pp. 149-155, 2020.
[CrossRef] [Google Scholar] [Publisher Link]
[8] Chen-Han Tsai et al., “Knee Injury Detection Using MRI with Efficiently-Layered Network (ELNet),” Proceedings of Machine Learning Research, vol. 121, pp. 784-794, 2020.
[Google Scholar] [Publisher Link]
[9] Sushovan Chaudhury et al., “A Novel Approach to Classifying Breast Cancer Histopathology Biopsy Images Using Bilateral Knowledge Distillation and Label Smoothing Regularization,” Computational and Mathematical Methods in Medicine, vol. 2021, pp. 1-11, 2021.
[CrossRef] [Google Scholar] [Publisher Link]
[10] Stanford University, MRNet: A Dataset of Knee MRIs and Competition for Automated Knee MRI Interpretation Stanford University. Serra Mall, Stanford, 2018. [Online]. Available: https://stanfordmlgroup.github.io/competitions/mrnet/
[11] The LibreTexts Libraries, Anatomical Position and Planes. [Online]. Available:
https://bio.libretexts.org/Bookshelves/Human_Biology/Human_Anatomy_Lab/01%3A_Overview_and_the_Microscope/1.02%3A_An atomical_Position_and_Planes.
[12] Steven Paul Arnoczk, “Anatomy of the Anterior Cruciate Ligament,” Clinical Orthopaedics and Related Research, vol. 172, pp. 19- 25, 1983.
[Google Scholar] [Publisher Link]
[13] Imagenet, An Update to the ImageNet Website and Dataset, 2021. [Online]. Available: https://www.image-net.org/update-mar-11- 2021.php
[14] Shilpa Sharma et al., “A ResNet50-Based Approach to Detect Multiple Types of Knee Tears Using MRIs,” Mathematical Problems in Engineering, vol. 2022, pp. 1-9, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[15] Naveen Subhas et al., “Value of Knee MRI in the Diagnosis and Management of Knee Disorders,” Orthopedics, vol. 37, no. 2, pp. 109-116, 2016.
[CrossRef] [Google Scholar] [Publisher Link]
[16] Kavitha Joshi, and K. Suganthi, “Anterior Cruciate Ligament Tear Detection Based on Deep Convolutional Neural Network,” Diagnostics, vol. 12, no. 10, pp. 1-12, 2022.
[CrossRef] [Google Scholar] [Publisher Link]
[17] Fang Liu et al., “Deep Learning MR Imaging-Based Attenuation Correction for PET/MR Imaging,” Radiology, vol. 286, no. 2, pp. 676-684, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[18] Fang Liu et al., “Deep Learning Approach for Evaluating Knee MR Images: Achieving High Diagnostic Performance for Cartilage Lesion Detection,” Radiology, vol. 289, no. 1, pp. 160-169, 2018.
[CrossRef] [Google Scholar] [Publisher Link]
[19] Kaiming He et al., “Deep Residual Learning for Image Recognition,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, pp. 770-778, 2016.
[CrossRef] [Google Scholar] [Publisher Link]