Improving Robustness of Deep Learning Based Knee MRI Segmentation: Mixup and Adversarial Domain Adaptation

Abstract

Degeneration of articular cartilage (AC) is actively studied in knee osteoarthritis (OA) research via magnetic resonance imaging (MRI). Segmentation of AC tissues from MRI data is an essential step in quantification of their damage. Deep learning (DL) based methods have shown potential in this realm and are the current state-of-the-art, however, their robustness to heterogeneity of MRI acquisition settings remains an open problem. In this study, we investigated two modern regularization techniques -mixup and adversarial unsupervised domain adaptation (UDA) -to improve the robustness of DL-based knee cartilage segmentation to new MRI acquisition settings. Our validation setup included two datasets produced by different MRI scanners and using distinct data acquisition protocols. We assessed the robustness of automatic segmentation by comparing mixup and UDA approaches to a strong baseline method at different OA severity stages and, additionally, in relation to anatomical locations. Our results showed that for moderate changes in knee MRI data acquisition settings both approaches may provide notable improvements in the robustness, which are consistent for all stages of the disease and affect the clinically important areas of the knee joint. However, mixup may be considered as a recommended approach, since it is more computationally efficient and does not require additional data from the target acquisition setup.

Emergence of Deep Learning in Knee Osteoarthritis Diagnosis

“…To overcome the problem, the availability of public databases such as OAI, which has been widely used, contains enormous datasets that are suitable for future studies to compare their models. However, to ensure general applicability of the model, it is encouraged to adopt images from independent datasets to be included in the testing dataset of any DL model [ 36 ].…”

“…There are several works that have made integrations or extensions to improve the existing models. Panfilov et al [ 36 ] applied two regularization techniques, supervised mixup and unsupervised domain adaptation (UDA), to enhance the existing U-Net model on segmentation of articular cartilage and menisci. However, mixed results have been reported where mixup with weight decay potentially improves DSC performance, but UDA is relatively undesirable due to its heavy cost of computation.…”

Medical Image Segmentation With Limited Supervision: A Review of Deep Network Models

“…Mixing and cutting images [67]- [69], [80] is a class of simple but effective augmentation methods in many applications [70]. Specifically, Mixup [67] linearly interpolates a random pair of training images and correspondingly their labels.…”

“…Alternatively, CutMix [68] is a combination of aspects of Mixup and Cutout by replacing a portion of an image with a portion of a different image. For the application of medical image segmentation, Panfilov et al [70] tested the efficiency of Mixup for knee MRI segmentation and showed improved model robustness.…”

Applications of machine learning algorithms in quality of life

“…Flexible seeds labelling applied on MRI data [129] was the dominant approach on the integrative segmentation category. To enable automation on the segmentation tasks, advanced DL-based techniques were adopted (e.g., CNN [130][131][132], unsupervised domain adaptation DL [133] and DNN [134] or even state-of-the-art ML techniques such as SVM [135], KNN [136,137] and RF [138,139]). Finally, more traditional segmentation approached were also proposed including: two-pass block discovery mechanism [140], Iterative Local Branch-andmincut [141], Gaussian fit model [142] and multi-atlas segmentation (MAS) [143].…”