Abstract Alzheimer's disease (AD) is the most prevalent form of dementia, defined as a biological aging process that can be detected through specific biomarkers even in the absence of typical symptoms, closely associated with mitochondrial dysfunction. However, there is still a lack of research on convenient and non-invasive mitochondrial biomarkers. This study developed an artificial intelligence model that using live urine-derived stem cell (USC) mitochondrial fluorescence imaging to effectively differentiate between individuals with cognitive impairments (AD) and mild cognitive impairment (MCI)), and cognitively normal (CN) subjects. Initially, we performed segmentation on the mitochondrial fluorescence images of living Hela cells and trained two binary classification models using the ResNet18 convolutional neural network. These models were designed to identify mitochondrial hyperfission versus normal , and mitochondrial hyperfusion versus normal. The models demonstrated robust performance in identifying intermediate mitochondrial states during testing. Ultimately, this model successfully recognized the USC mitochondria in individuals with cognitive impairment. Overall, our findings suggest that this model holds promise as an effective tool for the early diagnosis of Alzheimer's disease.
This repository can be used for training and inferring on mitochondrial fluorescence images.
pip install pytorch torchvision
pip install tensorboard
pip install scikit-learn
pip install opencv
Below are the main directories in the repository:
data_pre/: We segment the mito fluorescence images and crop the regions of interest into patches of size 112*112scripts/: scripts for training and evaluating the models
# extract images from lif files
python data_pre/main_crop_images_from_lif.py
# Clean the masks and select the ones with the highest possibility of being mitochondria
python data_pre/mask_clean_step_0_new.py
# Crop the regions of interest into patches of size 112*112 and select the best patches
python data_pre/patch_selection_step_1.py
# Train the classification model
python scripts/resnet_patch_three_dist_step_3.py
# Test the classification model and evaluate the results
python scripts/resnet_patch_three_dist_test_step_4.py
# Evaluate the results of the classification model on other datasets
python scripts/resnet_patch_three_eval_others_step_5.py