Automatic Grading of Knee Osteoarthritis from Plain Radiographs Using Densely Connected Convolutional Networks

Abstract

In this paper, we consider densely connected convolutional networks and their applicability to the problem of assessment of knee osteoarthritis (OA) severity in the five-point Kellgren-Lawrence scale. First, we use trained from scratch Single Shot Detector (SSD) to localize knee joint areas in radiographs. Then, we apply DenseNets to quantify OA stages in the images of detected knee joints. We consider networks of different depths, trained both from scratch and pre-trained on the ImageNet dataset and fine-tuned in the images from Osteoarthritis Initiative dataset (OAI). Also, different loss functions are examined to understand which one gives the best training results. In the knee joint localization task, we obtain an accuracy of 94.03% under the Jaccard index threshold of 0.75. Also, our classifier outperforms the current state-of-the-art with accuracy of 71% in the classification task.

Automating classification of osteoarthritis according to Kellgren-Lawrence in the knee using deep learning in an unfiltered adult population

“…Tiulpin et al [ 23 ] recently showed a new DL approach to significantly increase detection of radiographic OA presence. Mikhaylichenko et al [ 12 ] paper have also shown promising results in applying different types of architecture in tackling knee OA grading, using the KL grading scale. Like our study, the lowest accuracy was in the mid-categories of the KL grading scale.…”

“…Major differences between our study and that of the above-mentioned ones are how we let the network learn from an entire image series without any preprocessing. Comparable studies often use big ImageNet datasets like that of “The Osteoarthritis Initiative (OAI)” [ 12 , 15 ] or “Multicenter Osteoarthritis Study (MOST)” [ 11 , 23 ] to train their networks. In contrast, all images used in our training set was taken from a general setting where radiographs can differ greatly in quality and even beam angels.…”

Automated knee osteoarthritis severity classification using three‐stage preprocessing method andVGG16architecture

“…The fine‐tuned model achieved a mean absolute error value of 0.28. Mikhaylichenko et al, 33 used a single shot detector (powered by MobileNet) to localize the knee joint in the radiograph image as a pre‐processing step. For classification, the authors used an ensemble of three DenseNet‐121.…”

Using squeeze-and-excitation blocks to improve an accuracy of automatically grading knee osteoarthritis severity using convolutional neural networks

“…В [7] проводится исследование сетей с архитектурой DenseNet различной глубины и с различными функциями потерь. Показано, что наилучшая среднеклассовая точность 68,98 % была достигнута для случая предтренированной сверточной сети архитектуры DenseNet-121 при использовании в качестве функции потерь классической кросс-энтропии.…”

“…Ввиду того, что в наборе данных отсутствует информация о позиционировании на рентгенограмме области сустава, для локализации этой области использовался готовый модуль локализации, также основанный на базе сверточной нейронной сети, описание которого можно найти в [7].…”