Annonce

Les commentaires sont clos.

Explainable deep learning for Mild Cognitive Impairment detection win the spectroscopy data

5 December 2022


Catégorie : Stagiaire


Context :

Alzheimer’s Disease (AD) is the most comment form of dementia. Neuroimaging data is an integral part of the clinical assessment providing a way for clinicians to detect brain abnormalities for AD diagnosis. Patients with AD suffer from the cognitive decline that leads to brain neurons and synaptic loss (i.e., memory loss, difficulty with problem-solving, etc.). Although there is currently no cure for AD, there are available medications that can slow down disease progression and improve the patient lifestyle. Recent studies on bio-markers research have demonstrated that the AD pathology is now suspected to start a long time before the manifestation of the clinical symptoms and even before brain damage. Hence, diagnosis of AD at earlier stages is of great clinical importance so that cognitive functions would be improved by medications and the spread of the disease would be prevented. Mild Cognitive Impairment (MCI) is an intermediary stage condition between healthy people and AD.

Detecting MCI subjects provide a potential window for early AD detection. However, MCI subjects’ detection remain a challenging clinical problem as it lies on a spectrum between NC and manifest AD. Therefore, identifying efficient bio-markers for early AD stages detection helps in establishing diagnosis and treatment strategies without delay. Over the last decades, imaging bio makers derived from anatomical Structural with machine learning techniques has been widely studied to assess brain atrophy for AD detection and prediction [1]. In addition to structural changes, metabolic changes in some brain regions could be a good biomarker for early AD detection [2]. However, the structural brain atrophy is not detectable at an early stage of the disease (namely for Mild Cognitive Impairment (MCI) and Mild Alzheimer's Disease (MAD). Indeed, potential biological bio-markers have been proved their ability to early detect brain abnormalities related to AD before brain structural damage and clinical manifestation. Magnetic Resonance Spectroscopy (MRS) is a non invasive technique providing a complementary approach to brain metabolism in vivo, during conventional MRI examination. MRS provides biological information of brain tissues at the molecular level allowing detecting brain abnormalities while MRI remains normal.

 

Objectives:

The goal of this internship is to:

  • develop new deep learning based models for spectroscopy data classification for early AD detection, namely the MCI class detection.
  • propose and implement a method for 1D Class Activation Map (CAM) generation for the 1D spectroscopy data for model interpretation. This task will the of a recently achieved work in our team [3]. The obtained 1D CAM should highlight the contributions of different MRS metabolites in the classification tasks. Data used in this internship are provided by CHU of Poitiers. In addition to the on MRS data, this data set contains multi-modal data of patients affected by different stages of AD (healthy elderly subjects, Mild Cognitive Impairment (MCI) and AD subjects)

Possibility to continue with a PhD proposal (starting in Seotember/October 2023) in Artificial intelligence for medical images analysis

Location : XLIM (Site de Futuroscope), université de Poitiers in collaboration with the CHU of Poitiers

References:

[1] Olfa Ben Ahmed et al "Recognition of Alzheimer's Disease and Mild Cognitive Impairment with multimodal image-derived biomarkers and Multiple Kernel Learning", International Journal Neurocomputing, vol. 220, p. 98-110, Elsevier 2017

[2] Wang Z, Zhao C, Yu L, et al Regional metabolic changes in the hippocampus and posterior cingulated area detected with 3-Tesla magnetic resonance spectroscopy in patients with mild cognitive impairment and Alzheimer's disease. Acta Radiol 2009; 50:312–19

[3] Kherchouche, Anouar, Olfa Ben Ahmed, Carole Guillevin, Benoit Tremblais, Adrien Julian, and Rémy Guillevin. "MRS-XNet: An Explainable One-dimensional Deep Neural Network for Magnetic Spectroscopic Data classification." In The 29th IEEE International Conference on Image Processing (IEEE ICIP) 2022. 2022

Requirements :

Master 2 in computer vision, image processing, machine learning or any related field

Strong programming skills in python and deep learning frameworks (tensorflow, pytorch)

Tenattive Start date February/march 2023

Duration 5/6 months

Application : Send CV + transcripts and 2 reference letters to olfa.ben.ahmed@univ-poitiers.fr