3d Unsupervised Kidney Graft Segmentation Based On Deep Learning And Multi-Sequence Mri

Abstract

Image segmentation is one of the most popular problems in medical image analysis. Recently, with the success of deep neural networks, these powerful methods provide state of the art performance on various segmentation tasks. However, one of the main challenges relies on the high number of annotations that they need to be trained, which is crucial in medical applications. In this paper, we propose an unsupervised method based on deep learning for the segmentation of kidney grafts. Our method is composed of two different stages, the detection of the area of interest and the segmentation model that is able, through an iterative process, to provide accurate kidney draft segmentation without the need for annotations. The proposed framework works in the 3D space to explore all the available information and extract meaningful representations from Dynamic Contrast-Enhanced and T2 MRI sequences. Our method reports a dice of 89.8 ± 3.1%, Hausdorff distance at percentile 95% of 5.8±0.41mm and percentage of kidney volume difference of 5.9 ± 5.7% on a test dataset of 29 patients subject to a kidney transplant.

Level-Set-Based Kidney Segmentation from DCE-MRI Using Fuzzy Clustering with Population-Based and Subject-Specific Shape Statistics

“…In recent years, CNN in general, and the U-Net architecture in particular [ 13 ], have been applied to various medical image segmentation problems with good results [ 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 ]. Therefore, we have compared the proposed method versus a base U-Net CNN and one of its variants named BCDU-Net [ 28 ].…”

“…Haghighi et al [ 12 ] employed two cascaded U-Net models [ 13 ] to segment kidneys from 4D DCE-MRI data. Later on, Milecki et al [ 14 ] developed a 3D unsupervised CNN-based approach for the same reason. Bevilacqua et al [ 15 ] presented two different CNN-based approaches for accurate kidney segmentation from MRI data.…”

Contrastive Masked Transformers for Forecasting Renal Transplant Function

“…First simple statistics from the serum creatinine captured from the available blood test results between each follow-up (number of points, mean, median, standard deviation, minimum, maximum) are calculated and used as input to the models. Second, a set of predefined radiomics features [10] are obtained from the segmentation of the kidney transplant following the unsupervised method presented in [17]. Finally, we investigate generating MRI features from SimCLR [4] contrastive scheme, while we report the performance of different transfer-learning approaches, pre-trained on ImageNet [6] by duplicating the weights to 3D, Kinetics [23], and medical image datasets MedicalNet [3].…”

“…The DCE MRI volumes sized 512 × 512 × [64 − 88] voxels included spacing ranging in [0.78 − 0.94] × [0.78 − 0.94] × [1.9 − 2.5] mm. All volumes were cropped around the transplant using an automatic and unsupervised method for selecting the region of interest and reducing dimensionality (Milecki et al, 2021). Intensity normalization was executed to each volume independently by applying standard normalization, clipping values to [−5, 5], and rescaling linearly to [0, 1].…”

“…In [21], diverse imaging modalities have been investigated to assess renal transplant functions in several studies. Moreover, in [17] multiple Magnetic Resonance Imaging (MRI) modalities are used for the unsupervised kidney graft segmentation. Beyond the respective limitations of the several imaging modalities, such as the necessity of radiations or the intrinsic trade-off on resolution, to our knowledge, there are no studies focusing on monitoring the evolution of kidney grafts using imaging data.…”