HOPE-for-mild-cognitive-impairment

[JBHI 2024] This is a code implementation of the hybrid-granularity ordinal learning proposed in the manuscript "HOPE: Hybrid-granularity Ordinal Prototype Learning for Progression Prediction of Mild Cognitive Impairment". [doi] [arxiv]

Introduction

Existing works typically require MCI subtype labels—progresive MCI (pMCI) vs. stable MCI (sMCI)—determined by whether or not an MCI patient will progress to AD after a period of follow-up diagnosis. However, collecting retrospective MCI subtype data is time-consuming and resource-intensive, which leads to relatively small labeled datasets, resulting in amplified overfitting and challenges in extracting discriminative information.

Hybrid-granularity ordinal prototype learning

Based on the ordinal development of AD, we take a fresh perspective on the extensive cross-sectional data collected from subjects across all stages of AD, ranging from Normal Cognition (NC) to MCI to AD, as the ''latent'' longitudinal data specific to the entire AD cohort; the pathological differences between sMCI and pMCI are analogical to those between NC and AD. Inspired by this, we propose a novel Hybrid-granularity Ordinal PrototypE learning (HOPE) method to predict the progression of MCI by learning the ordinal nature of AD.

Experimental results

Experimental results on the internal ADNI and external NACC datasets show that the proposed HOPE outperforms recently published AD-related and ordinal-based state-of-the-art methods and has better generalizability.

Moreover, we present data visualization using GradCAM and t-SNE. Our findings indicate that our HOPE has effectively learned the ordinal nature of AD development. Furthermore, we have identified specific regions of interest that are closely associated with the progression of AD.

Usage

Install requirements

First clone the repository:

git clone https://github.com/thibault-wch/HOPE-for-mild-cognitive-impairment.git

And then install other requirements:

pip install -r requirements.txt

Data prepration

We trained, validated and tested our HOPE using the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. To investigate the generalizability of the proposed HOPE, we externally tested our HOPE on the National Alzheimer's Coordinating Center (NACC) dataset.

We first use Freesurfer and Flirt for preprocessing all MRI images. Furthermore, we transform the preprocessed .nii.gz brain files into .npy format. In addition, we reorganize and split them into the data pairs for ./utils/Dataset.py using pickle, the concrete data pair demo as shown in:

{   
    '0': [
    (  # the MRI file path
        '/data/chwang/final_dataset_MRI/lineared/train/0_141_S_0810.npy',
        # the diagnosis label of the corresponding MRI subject (NC->0 AD->1 sMCI->3 pMCI->4)
        0),
          ...],
    '1': [
    (   '/data/chwang/final_dataset_MRI/lineared/train/1_137_S_0841.npy',
        1),
          ...],
    '3': [...],
    '4': [...]
}

Implementation details

We implement all the methods with the PyTorch library and train the networks on NVIDIA V100 GPUs. All networks are built with 3D ResNet18 as the backbone, initialized by the Kaiming method and trained 60 epochs using the Adam optimizer with $\beta_1 = 0.5$ and $\beta_2 = 0.999$. We set the initial learning rate to $2\times 10^{-4}$ and then gradually reduce it using exponential decay with a decay rate of 0.95. The batch size is set to 8. We have organized the concrete training and inference process in ./scripts.

For Training

cd ./scripts
sh train.sh

For Inference

cd ./scripts
sh test.sh

Folder structure

HOPE-for-mild-cognitive-impairment
  ├─ models
  │   ├─ Resnet
  │   ├─ ranking <ranking components>
  │   └─ BasicComputing <instance-to-class, class-to-class components>
  ├─ options (different options)
  ├─ scripts (different phases' scripts)
  ├─ utils
  │   ├─ Dataset <our defined dataset>
  │   ├─ train_data <training step>
  │   ├─ test_data <testing step>
  │   └─ ...
  ├─ readme_files
  ├─README.md
  ├─requirements.txt
  ├─train
  └─test

Acknowledgement

• We gratefully thank the ADNI and NACC investigators for providing access to the data.

• Our code is inspired by Blackbox Combinatorial Solvers and RankSim.

Citation

If you find this work useful for your research, please 🌟 our project and cite our paper :

@article{wang2024hope,
title = {HOPE: Hybrid-granularity Ordinal Prototype Learning for Progression Prediction of Mild Cognitive Impairment}, 
author = {Chenhui Wang and Yiming Lei and Tao Chen and Junping Zhang and Yuxin Li and Hongming Shan and others},
year = {2024},
journal={IEEE Journal of Biomedical and Health Informatics},
}