Open matteofrigo opened 5 years ago
Very interesting project!! You might want to take a look at neuropredict to help with analyzing the predictive utility of various features : github.com/raamana
Nice work! I can participate from remote
Thanks @raamana for the suggestion!
@maurozucchelli thank you for your help! We'll set up a way to let you (and possibly others) to join us remotely.
Hi @matteofrigo is this still open? I would like to work on the ADNI 3 data but can't find the multi-shell data on their repository. Thanks
Great project ! Can I ask : what are the limitations for re-sharing original and derived data from ADNI data usage agreement?
JB, ADNI's DUA (like ABCD and some others) simply prohibits from redistributing anything (raw data, or derived features) outside a collaboration for which the DUA was signed for.. so they can never post the features or raw images online. They could share it with internal collaborators at the same institute..
I wrote to NIH about it, but they were discouraging (and even intimidated me to not pursue the change of policies).
Wouldnt this be an issue for this project ?
Not for me as I don't need the features shared to me -- I was mostly thinking advising them (on ML, biomarker performance eval and neuropredict etc) at that time (May 2019).. Not sure where the project stands, and what their plans are now..
Hi @jbpoline and @raamana , I reported your concerns to my collaborators that work in ADNI, which in turn said that they would discuss this issue with the PIs.
Regarding this specific project, we did a preliminary analysis and we presented it last year at ISMRM 2020. Hopefully in the future we will find some time to dive deeper into the problem!
Tracking microstructural biomarkers of Alzheimer’s Disease via diffusion MRI
The Alzheimer’s Disease Neuroimaging Initiative (ADNI) is a longitudinal natural history study. It is a large multicenter study designed to identify clinical, MRI, genetic, and biochemical markers for the early detection and tracking of Alzheimer's disease (AD). In particular, identifying biomarkers sensitive to mild cognitive impairment (MCI) is important to better categorize the transitional stages between normal aging and AD, and to evaluate targeted treatments.
Data from ADNI is publicly available. The third phase of ADNI (ADNI-3) began in late 2016, with subject imaging beginning in mid-2017. ADNI-3 includes an advanced multi-shell diffusion MRI acquisition, besides the basic single-shell acquisition [1] (see Figure 1). Multi-shell dMRI allows for the reconstruction of diffusion models beyond Diffusion Tensor Imaging (DTI).
ADNI-3 Advanced multi-shell protocol:
Figure 1. Comparison of “basic” and “advanced” diffusion MRI protocols in ADNI-3. Taken from Reid et al. 2017 [1].
In multi-shell data, multi-compartment models can be used to delineate the signal contributions of different tissue compartments, which in turn tell us something about the tissue’s microstructural composition. Conveniently, Dmipy is an open source tool designed to modularly generate and fit any state-of-the-art multi-compartment diffusion models on-the-fly. Here, we aim at fitting all possible multi-shell models for the ADNI3 advanced diffusion protocol with Dmipy and benchmark which model is best to be used as an imaging biomarker to track the progression of Alzheimer’s Disease in the elderly.
Multi-compartment models that are relevant for multi-shell microstructure exploration are: Ball and Stick [2], NODDI-Watson [3], NODDI-Bingham [4], Multi-compartment microscopic diffusion imaging (MC-MDI) [5] and Multi-Tissue CSD [6]. Aside from parametric models, we also evaluate if signal-based markers from signal models such as MAP-MRI [7] can be valuable markers for tracking AD (RTOP, RTAP, RTPP, MSD, NG).
The aim of this project is to determine the best diffusion model (if any) to measure the intra-cellular, extracellular volume fractions, and the dispersion of fibers, whose change should correlate with the pathological progression of AD.
Comparison across dMRI models
For each dMRI measure, we will run a logistic regression with TV-L1 regularization (Nilearn package) across voxels to classify individuals with mild cognitive impairment (MCI; N=17; mean age: 76.8±7.5 yrs; 14M/3F) from those who are cognitively normal (CN; N=39; mean age: 73.2±7.2 yrs; 25M/14F) to identify which dMRI measure gives the highest classification accuracy. Among dMRI measures yielding >80% accuracy we will compare the Jaccard/Dice similarity coefficient from the resulting maps of classifying regions to identify which dMRI measures give similar information in similar regions and which offer additional information about underlying pathological changes.
Note
We may use different classification labels between groups, which can be based on commonly used screening tools for detecting dementia and AD as the Alzheimer’s Disease Assessment Scale 13 (ADAS-cog8), the Mini-Mental State Examination (MMSE9), and the Clinical Dementia Rating scale sum-of-boxes (CDR-sob10), amyloid PET scores, or cerebrospinal fluid (CSF) markers.
Skills required to participate
We welcome any curious brainhacker who is interested in improving the understanding of the Alzheimer's Disease and/or wants to see how simple it can be to study tissue microstructure with python.
Integration
The goal is to track the changes of tissue microstructure in AD. Ideally, we will find a microstructural biomarker that lets us anticipate the classical symptoms of AD, giving us the possibility to set up the corresponding therapy in advance. We will be analyzing many different models for each subject; this will raise problems related to dimensionality reduction and feature selection.
Your collaboration will be precious in:
Preparation material
You can have a look at the website of ADNI to get know more about the data we are processing. To get informations about the fitting of tissue microstructure models, you can look at the website of Dmipy.
Links
Communication
This issue will be kept as reference discussion channel. Question can also be directly addressed to @villalonreina (ADNI) and @rutgerfick (Dmipy).
References