Deep Learning with Imbalanced Data Identifies Early Mild Cognitive Impairment Amongst Typical Older Adults (Preprint)

Author:

Zapata Ruben DarioORCID,Albizu Alejandro,Perrin Cameron,O‘Shea Andrew,Hardcastle Cheshire,Kraft Jessica,Alexander Gene E.,Marsiske Michael,Cohen Ronald A.,Woods Adam J.,Gullett Joseph M.ORCID

Abstract

BACKGROUND

Mild cognitive impairment is a transitional stage between normal cognitive aging and dementia. Identifying individuals at the preMCI stage, prior to the onset of mild cognitive decline, can be pivotal for early interventions aimed to reduce the progression neurodegeneration.

OBJECTIVE

The objective of this study is to develop convolutional neural networks trained on fluid attenuated inversion recovery magnetic resonance imaging for classification of preMCI in an imbalanced cohort of 350 participants.

METHODS

A DenseNet 264 convolutional neural network was trained on an imbalanced dataset of 350 participants with a dataset split into 64%, training, 16% validation, and 20% testing sets. Training was conducted with a batch size=70, epoch=200, processing images resized to a uniform dimension of 128×128×128 voxels, and optimizer=Adam. The optimization of our network was conducted using the Adam Optimizer with a learning rate of 10-3 and a weight decay of 5-4. Data augmentation strategy included Random affine transformation, random rotation across an axis, and random Gaussian noise on each image during training.

RESULTS

Mean age of the participants was 71.6 (SD 5.14); the average educational attainment was 16.3 years (SD 2.39). The mean MoCA score was 26.7 (SD 1.93). Our DenseNet mode achieved R2=0.146 and RMSE=0.569.

CONCLUSIONS

These findings underscore the potential of brain imaging and DL to identify this at-risk population, offering a promising tool for early detection and potential personalized preventative strategies against cognitive decline. Further research is warranted to improve upon the results to validate these findings in a larger, more diverse population.

Publisher

JMIR Publications Inc.

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3