Learning From Mouse CT-Scan Brain Images To Detect MRA-TOF Human Vasculatures

Abstract

The earlier studies on brain vasculature semantic segmentation used classical image analysis methods to extract the vascular tree from images. Nowadays, deep learning methods are widely exploited for various image analysis tasks. One of the strong restrictions when dealing with neural networks in the framework of semantic segmentation is the need to dispose of a ground truth segmentation dataset, on which the task will be learned. It may be cumbersome to manually segment the arteries in a 3D volumes (MRA-TOF typically). In this work, we aim to tackle the vascular tree segmentation from a new perspective. Our objective is to build an image dataset from mouse vasculatures acquired using CT-Scans, and enhance these vasculatures in such a way to precisely mimic the statistical properties of the human brain. The segmentation of mouse images is easily automatized thanks to their specific acquisition modality. Thus, such a framework allows to generate the data necessary for the training of a Convolutional Neural Networki.e. the enhanced mouse images and there corresponding ground truth segmentation -without requiring any manual segmentation procedure. However, in order to generate an image dataset having consistent properties (strong resemblance with MRA images), we have to ensure that the statistical properties of the enhanced mouse images do match correctly the human MRA acquisitions. In this work, we evaluate at length the similarities between the human arteries as acquired on MRA-TOF and the "humanized" mouse arteries produced by our model. Finally, once the model duly validated, we experiment its applicability with a Convolutional Neural Network.

Toward a 3D Arterial Tree Bifurcation Model for Intra-Cranial Aneurysm Detection and Segmentation

“…In order to evaluate the applicability of our model within a Deep Neural Network, we have set up a 3D U-Net to perform the segmentation of small 3D cropped bifurcations. The network was first trained on 120 modeled bifurcations and provided an average DICE score of 0.8 on a test set composed of 41 images, which is fairly close to the performances achieved in [15]. Next, to evaluate the model's suitability as a generator of augmented data, we have first launched the U-Net on 120 ground truth bifurcations, and then, on a set composed of both 40 ground truth and 80 modeled bifurcations.…”

“…Hopefully, this drift might not be strong enough to have an impact on the upcoming use of this dataset for Deep Learning ICA detection. In previous works [15], we managed to efficiently train a CNN (Half U-Net) using only modeled arteries, we believe that despite the more sophisticated model presented here (3D artery and ICA), such a training scenario should still be feasible. Moreover, the model might also be used as backup to some ground truth segmentations to train a neural network (acting as augmented data).…”

“…We propose to build up a full model of both the 3D bifurcation and the aneurysm, in order to train a deep learning network to recognize the ICA. Our approach here is somewhat similar to a previous project we have recently been investigating [15]. Nevertheless, the approach presented here is much more advanced, with regard to several aspects: i) in [15], a 2D artery dataset was built, whereas here, we aim to tackle not only the artery, but rather the whole bifurcation in 3D; ii) the bifurcation bears an unruptured aneurysm; and iii), the background noise is now characterized and modeled with a much better accuracy.…”

“…Our approach here is somewhat similar to a previous project we have recently been investigating [15]. Nevertheless, the approach presented here is much more advanced, with regard to several aspects: i) in [15], a 2D artery dataset was built, whereas here, we aim to tackle not only the artery, but rather the whole bifurcation in 3D; ii) the bifurcation bears an unruptured aneurysm; and iii), the background noise is now characterized and modeled with a much better accuracy. However, it is important to notice that this current work is only devoted to the design of the model itself.…”

How feasible is it to segment human glomerulus with a model trained on mouse histology images?

“…Although there is some works using mouse and human species outside the domain of histology [Chater et al, 2021, Hossain et al, 2021, only a few deep learning studies on histological images have been exploring samples of these species. Bouteldja et al [2021] developed a custom U-Net network for automated multi-class segmentation of glomerular images of different mammalian species, not only mice and humans.…”