Browsing by Author "CHERIFI, Dalila (Supervisor)"

Now showing 1 - 2 of 2

Melanoma identification using convolutional neural networks
(2018) Louifi, Akram; Soulami, Ameur; CHERIFI, Dalila (Supervisor)
Melanoma is an extremely dangerous type of skin cancer causing fatal incidences, it’s also an increasing form of cancer around the world. Since the odds of recovering for the earlydiagnosed cases is very high, early detection of melanoma is vital. Computer assisted diagnosis have been used alongside traditional techniques so as to improve the reliability of detecting melanoma. In this project, a convolutional Neural network model designed from scratch as well as Transfer Learning using the pretrained model Inception v3 are used in order to develop a reliable tool able to detect melanoma that can used by clinicians and individual users. The results using Inception v3 model for dermoscopical images achieved the best results compared to our model. The results are compared to those of clinicians, which shows that the algorithms can be used reliably for the detection of melanoma.
Tractography of White Matter Fibers in Presence of Astrocytoma
(2018) BENOUADAH, Sara; ESSAHELI, Mohamed Abderaouf; CHERIFI, Dalila (Supervisor)
Diﬀusion Magnetic Resonance Imaging (dMRI), a technique that maps the axonal mi- crostructure of the brain, is often used nowadays to investigate white matter alterations. Clinicians have gained useful insights from these studies for surgical planning and demon- strating subtle abnormalities in a variety of diseases. Our work aims to identify the eﬀects of astrocytoma, a type of tumor that aﬀects the brain, on white matter (WM) tracts. For this purpose, three patients with diﬀerent grades of astrocytomas, acquired by the UK Data Archive, are used. Constrained Spherical De- convolution (CSD)-based deterministic tractography is applied on these datasets in order to assess the tumor-induced alterations on WM tracts. Three types of these alterations were observed: displacement of ﬁbers around the tumor, destruction of the intralesional ﬁbers, and destruction with displacement of the ﬁbers within the tumoral region. A comparison is also performed between the results obtained using diﬀerent reconstruction methods: Diﬀu- sion Tensor Imaging (DTI) and CSD. Although both methods conﬁrm the existence of the tumor and its associated alterations, CSD-based tractography is most appropriate to use when dealing with tasks sensitive to intravoxel ﬁber connections such as the planning of a surgical procedure. Finally, a comparison between the left and right hemispheres reveals that the number of streamlines can be used to predict the existence of a tumor. The results were validated by a group of doctors and researchers in the ﬁeld. The main contributions of this work reside in determining a pattern to the eﬀects of astrocytomas on WM tracts. While monitoring the ﬁeld, we found that, among the re- searches investigating brain pathologies, none were particularly focused on this type of tumor. In addition, the use of CSD-based deterministic tractography while investigating brain pathologies is new in the literature, where most published articles use DTI as the only local reconstruction technique. Another contribution of this work is the comparison between the left and right hemispheres of both normal and brain tumor patients.