Segmentation of Multiple Sclerosis in MR images by Deep Learning
dc.contributor.author | BENHIZIA, Louiza | |
dc.contributor.author | BENBATATA, Sabrina | |
dc.date.accessioned | 2022-01-05T08:09:02Z | |
dc.date.available | 2022-01-05T08:09:02Z | |
dc.date.issued | 2021 | |
dc.description.abstract | Multiple sclerosis (MS) is one of the most difficult diseases to diagnose or follow-up, especially in the first stage. Magnetic resonance imaging shows the structure of the central nervous system which makes it possible to detect. However, The manual MS diagnosis is often painstaking and requires significant and tedious efforts. Also, MRI does not achieve the diagnosis without an expert neuroradiologist. In this work, we propose an automatic segmentation of MS from MRI using deep learning. Our proposed solution is using a segmentation method called Unet on the public MS dataset, the MR images used in this work are T2. Our proposal was validated in the dataset of MS against other work in terms of the dice similarity coefficient (DSC) metric. The proposed Unet model had an accuracy of 99%, which may compete with recent work. key-Words : Magnetic Resonance imaging, Segmentation, Multiple sclerosis, Unet. R´esum´e La scl´erose en plaques (SEP) est l’une des maladies les plus difficiles `a diagnostiquer, surtout au d´ebut. Alors que les images IRM du syst`eme nerveux central montrent la structure du tissu qui permet de le d´etecter. Le diagnostic manuel de la SEP est souvent laborieux et n´ecessite des efforts importants. De plus, l’IRM n’atteint pas le diagnostic sans un neuroradiologue expert. Dans ce travail, nous proposons une m´ethode de segmentation des images IRM de la SEP par l’apprentissage profond. La solution propos´ee utilise la m´ethode de segmentation appel´ee Unet sur l’ensemble des donn´ees SEP, les images IRM utilis´ees dans ce travail sont une T2. Notre proposition a ´et´e valid´ee dans la base de donn´ees de SEP contre d’autres travaux en termes de m´etrique du coefficient de similarit´e des d´es (DSC). Le mod`ele Unet propos´e avait une pr´ecision de 99%. Ce qui peut concurrencer les travaux r´ecents. Mots cl´es : Imagerie par r´esonnance magn´etique, Segmentation, Scl´erose en Plaques, Unet. P l ¨ T}A ¤ ,T` Atm ¤ Py Kt ¨ T w`} r ± |r ± d w¡ d`tm lOt Tyn rh\ ©z rm ¨bO` EAh l ¨syVAn m y r A r§wOt y ¨ .Y ¤± Tl rm lOt |rm ©¤dy Py Kt , Ð ¤ .¢ AKt kmm ` ¨t T s ± ¨syVAn m y r A Py Kt ¤ .¾®§wV A¾At ¤ r ts§¤ A¾ryb A¾dh lWt§ A A¾Ab A d`tm lOt ¨¶Aqlt Kk An rt , m` @¡ ¨ .ryb AO T`J ¨¶AO Y At § An§d rtqm . ym` l`t d tFA ¨syVAn m y r A r§wOt §rV d`tm y r Cw} , d`tm lOt A Ay T wm Yl Unet Yms ysq Tq§rV d tF w¡ A Ay T wm ¨ An rt Yl §dOt dq .T2 ¨¡ m` @¡ ¨ T d tsm ¨syVAn m rtqm Unet Ðwm .©rtm (DSC) ¢ AKt A` y «r Am dR d`tm lOt .«r ± Am ± H Ant d ¨t ¤ ,T¶Am A 99 T ¢§d | en_US |
dc.identifier.issn | MM/656 | |
dc.identifier.uri | http://10.10.1.6:4000/handle/123456789/1657 | |
dc.language.iso | en | en_US |
dc.publisher | Université Mohamed el-Bachir el-Ibrahimi Bordj Bou Arréridj Faculté de Mathématique et Informatique | en_US |
dc.subject | Magnetic Resonance imaging, Segmentation, Multiple sclerosis, Unet | en_US |
dc.title | Segmentation of Multiple Sclerosis in MR images by Deep Learning | en_US |
Files
Original bundle
1 - 1 of 1
- Name:
- memoire_master.pdf
- Size:
- 4.83 MB
- Format:
- Adobe Portable Document Format
- Description:
- Medical image segmentation remains a very vast area of research because there are many diseases which are developing every time or very difficult to detect them in medical images. And multiple sclerosis is one of those diseases. The manual segmentation of MS lesions by physicians consists of subjective and time consuming procedural sequences. For this reason, automatic detection of MS using MR images is important in both terms time consuming and cost. Recently, computer-aided tools assist physicians in detecting, diagnosing and following-up MS on magnetic resonance imaging MRI. The goal of this work has been segmenting multiple sclerosis from MR images using deep learning. Where, we first presented MS disease and medical imaging and we focused on MRI. After that, we made a literature review on image segmentation using deep learning methods. In our study we chose the Unet method in the public MS dataset for the segmentation of MS from MRI. The process of detecting MS has four fundamental phases, first we collected data (dataset). Then, we did a pretreatment and augment data with rotation, shift and horizontal flip. After that, we trained our dataset on 30 iterations. Finally, we showed the results and we evaluated our model. The results obtained after segmentation on the dataset are satisfying. Our work is only in its initial version. We can say that it is still open for comparison and/or hybridisation work with other segmentation methods and with other dataset
License bundle
1 - 1 of 1
No Thumbnail Available
- Name:
- license.txt
- Size:
- 1.71 KB
- Format:
- Item-specific license agreed to upon submission
- Description: