Knee-joint tissue recognition in magnetic resonance imaging

Suponenkovs, Artjoms; Markovičs, Zigurds; Platkājis, Ardis

doi:10.1109/nc.2017.8263280

Cited by 2 publications

(5 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As a non-invasive modality with high spatial resolution without ionizing radiation and multi-planar capability, MRI can capture useful structures of the joint, e.g., cartilage morphology and biochemical composition [40], [41]. On the other hand, MRI is very sensitive to full or by the success of HGE [10], [14], [47]- [49] [46]. Even though these approaches cut off details by smoothing, they are merely applicable for classifying the thickness of cartilages.…”

Section: B Magnetic Resonance Imagingmentioning

confidence: 99%

“…Even though these approaches cut off details by smoothing, they are merely applicable for classifying the thickness of cartilages. Moreover, since MRI struggles with diversiform pulse sequences and parameters, affine registration is applied to solve the multiformity of pulse sequences and parameters, outperforming Sobel filter [14].…”

Section: B Magnetic Resonance Imagingmentioning

confidence: 99%

“…Whole images are then decomposed into a set of cells, whose intensity is used as features for the classification with Gaussian mixture model (GMM). Through a genetic algorithm (GA), classified cells are iteratively grouped by a heuristic search, where K-means is adopted to control the number of clusters in case of creating many segments like Watershed algorithm [14].…”

Section: B Magnetic Resonance Imagingmentioning

confidence: 99%

“…Subsequently, several approaches have been proposed to quantify the OA severity based on the segmented image intensity. However, articular cartilage is not the only determinate factor [7] MLP Model trained on limited cohorts, i.e., 100 patients for the experiments Suponenkovs A. et al [14] K-means Used only to learn the pattern and for statistical analysis of selected features Kubakaddi S. et al [47] SVM Model trained on limited cohorts (15 patients) for the experiments Sharma G. et al [49] SVM Model trained on limited cohorts (16 patients) for anomaly detection in binary classification Huang C. et al [51] GHMM Diseased region detection focus…”

Section: B Magnetic Resonance Imagingmentioning

confidence: 99%

“…Therefore, each modality on its own may hence not contain sufficient information for an accurate OA quantification or diagnosis. Further, artifacts such as aliasing, slice overlap, truncation, patient motion, and pathological features like osteophytes, bone marrow lesions, and surface fibrillation also impede accurate diagnosis of OA [14]. Deep learning (DL) techniques propelled recent advancements in computer vision techniques which again lead to a rise in automated medical diagnoses solutions and computer-aided clinical decision-making [15].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

DeepKneeExplainer: Explainable Knee Osteoarthritis Diagnosis From Radiographs and Magnetic Resonance Imaging

et al. 2021

View full text Add to dashboard Cite

Osteoarthritis (OA) is a degenerative joint disease, which significantly affects middleaged and elderly people. Although primarily identified via hyaline cartilage change based on medical images, technical bottlenecks like noise, artifacts, and modality impose an enormous challenge on high-precision, objective, and efficient early quantification of OA. Owing to recent advancements, approaches based on neural networks (DNNs) have shown outstanding success in this application domain. However, due to nested non-linear and complex structures, DNNs are mostly opaque and perceived as black-box methods, which raises numerous legal and ethical concerns. Moreover, these approaches do not have the ability to provide the reasoning behind diagnosis decisions in the way humans would do, which poses an additional risk in the clinical setting. In this paper, we propose a novel explainable method for knee OA diagnosis based on radiographs and magnetic resonance imaging (MRI), which we called DeepKneeExplainer. First, we comprehensively preprocess MRIs and radiographs through the deep-stacked transformation technique against possible noises and artifacts that could contain unseen images for domain generalization. Then, we extract the region of interests (ROIs) by employing U-Net architecture with ResNet backbone. To classify the cohorts, we train DenseNet and VGG architectures on the extracted ROIs. Finally, we highlight classdiscriminating regions using gradient-guided class activation maps (Grad-CAM++) and layerwise relevance propagation (LRP), followed by providing human-interpretable explanations of the predictions. Comprehensive experiments based on the multicenter osteoarthritis study (MOST) cohorts, our approach yields up to 91% classification accuracy, outperforming comparable stateof-the-art approaches. We hope that our results will encourage medical researchers and developers to adopt explainable methods and DNN-based analytic pipelines towards an increasing acceptance and adoption of AI-assisted applications in the clinical practice for improved knee OA diagnoses.

show abstract