Introduction

This repository contains the experiments described in Supervised diagnosis prediction from cortical sulci : toward the discovery of neurodevelopmental biomarkers in mental disorders, presented at 21th IEEE ISBI 2024.

The paper can be found here.

Abstract: Recent advances in machine learning applied to structural magnetic resonance imaging (sMRI) may highlight abnormalities in brain anatomy associated with mental disorders. These disorders are multifactorial, resulting from a complex combination of neurodevelopmental and environmental factors. In particular, such factors are present in cortical sulci, whose shapes are determined very early in brain development and are a valuable proxy for capturing specifically the neurodevelopmental contribution of brain anatomy. This paper explores whether the shapes of cortical sulci can be used for diagnosis prediction using deep learning models. These models are applied to three mental disorders (autism spectrum disorder, bipolar disorder, and schizophrenia) in large multicentric datasets. We demonstrate that the neurodevelopmental underpinnings of these disorders can be captured with sMRI. Finally, we show the potential of visual explanations of models’ decisions in discovering biomarkers for mental disorders.

The goal is to predict diagnosis from cortical sulcus images for three mental disorders : * Healthy Control (HC) vs Autism Spectrum Disorder (ASD) * Healty Control (HC) vs Bipolar Disorder (BD) * Healthy Control (HC) vs Schizophrenia (SCZ)

Running experiments

Installation

To install the package, clone the repository into a folder and then : Shell cd /path/to/diag_pred_from_sulci pip install -e .

Datasets

The 3 clinical datasets SCZDataset, BDDataset and ASDDataset are derived mostly from public cohorts excepted for BIOBD, BSNIP1 and PRAGUE, that are private for clinical research. The three datasets are described below:

Dataset | # Subjects | Age (avg±std) | Sex (\%F) | # Sites | Studies | :---:| :---: | :---: | :---: | :---: | :---: | HC
SCZ | 761
532 | 33 ± 12
34 ± 12 | 51
29 | 12 | BSNIP1, CANDI, CNP, PRAGUE, SCHIZCONNECT HC
BD | 695
469 | 37 ± 14
39 ± 12 | 54
57 | 15 | BIOBD, BSNIP1, CANDI, CNP HC
ASD | 926
813 | 16 ± 9
16 ± 9 | 25
13 | 30 | ABIDE I , ABIDE II

To run experiments, you need a root folder containing : - the pickles of train-val-test schemes for each dataset - the mapping of acquisition sites - a folder morphologist with arrays of skeleton volumes and corresponding participant dataframes of each study

Launch model trainings

To launch model trainings, you need to launch the python script main.py in the dl_training folder. All the parameters to be passed into argument are explained in the script. ``` Shell python3 dl_training/main.py --args

if you need details about parameters

python3 dl_training/main.py --help ```

Experiments

1. Architecture selection : 3 CNN architectures have been tested, see the architecture folder
2. Loss selection : BCE and SupCon losses have been compared, see contrastive_learning folder for SupCon model
3. Pre-processing selection : Gaussian smoothing pre-processing, see img_preprocessing folder
4. XAI : an occlusion method have been applied to understand model decisions, see saliency_map folder

Results

You will find the prediction scores for the three mental disorders in the table below. Note that the results are averaged accross three trainings with different random initialization. Task | ROC AUC | | :---:| :---: | HC vs SCZ | 0.656 ± 0.034 | HC vs BD | 0.661 ± 0.038 | HC vs ASD | 0.595 ± 0.008 |

Citation

If you find this work useful for your research, please cite our paper: @inproceedings{Auriau:2024, author={Pierre Auriau and Antoine Grigis and Benoit Dufumier and Robin Louiset and Joel Chavas and Pietro Gori and Jean-François Mangin and Edouard Duchesnay}, title={Supervised diagnosis prediction from cortical sulci: toward the discovery of neurodevelopmental biomarkers in mental disorders}, booktitle={21st IEEE International Symposium on Biomedical Imaging (ISBI 2024)}, year={2024} }

Useful links

• Link to the archive : https://hal.science/hal-04494994
• First version of these scripts are at: https://github.com/Duplums/SMLvsDL
• More info of Brainvisa, software to extract cortical sulci from MRI images, at : https://brainvisa.info/web/