Predicting dementia progression with fully connected cascade neural networks

Akbarifar, Ahmad; Maghsoudpour, Adel; Mohammadian, Fatemeh; Mohammadzaheri, Morteza; Ghaemi, Omid

doi:10.21203/rs.3.rs-4115962/v1

Cited by 2 publications

(1 citation statement)

References 40 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By providing access to early diagnosis of dementia, this approach could pave the way for improved patient outcomes, reduced caregiver burden, and more efficient allocation of healthcare resources. 11 Several studies have investigated the potential of machine learning algorithms to predict dementia risk based on healthrelated factors. 12,13 Studies using variables, such as age, education, high blood pressure, obesity, and physical inactivity, have shown the possibility of identifying people at high risk of cognitive impairment.…”

Section: Introductionmentioning

confidence: 99%

Multimodal dementia identification using lifestyle and brain lesions, a machine learning approach

Akbarifar,

Maghsoudpour,

Mohammadian

et al. 2024

AIP Advances

Self Cite

View full text Add to dashboard Cite

Dementia diagnosis often relies on expensive and invasive neuroimaging techniques that limit access to early screening. This study proposes an innovative approach for facilitating early dementia screening by estimating diffusion tensor imaging (DTI) measures using accessible lifestyle and brain imaging factors. Conventional DTI analysis, though effective, is often hindered by high costs and limited accessibility. To address this challenge, fuzzy subtractive clustering identified 14 influential variables from the Lifestyle for Brain Health and Brain Atrophy and Lesion Index frameworks, encompassing demographics, medical conditions, lifestyle factors, and structural brain markers. A multilayer perceptron (MLP) neural network was developed using these selected variables to predict fractional anisotropy (FA), a DTI metric reflecting white matter integrity and cognitive function. The MLP model achieved promising results, with a mean squared error of 0.000 878 on the test set for FA prediction, demonstrating its potential for accurate DTI estimation without costly neuroimaging techniques. The FA values in the dataset ranged from 0 to 1, with higher values indicating greater white matter integrity. Thus, a mean squared error of 0.000 878 suggests that the model’s predictions were highly accurate compared to the observed FA values. This multifactorial approach aligns with the current understanding of dementia’s complex etiology influenced by various biological, environmental, and lifestyle factors. By integrating readily available data into a predictive model, this method enables widespread, cost-effective screening for early dementia risk assessment. The proposed accessible screening tool could facilitate timely interventions, preventive strategies, and efficient resource allocation in public health programs, ultimately improving patient outcomes and caregiver burden.

show abstract

Section: Introductionmentioning

confidence: 99%