TY - GEN
T1 - RWD184 FNEURONET
T2 - A SEX-SPECIFIC DEEP LEARNING FRAMEWORK FOR AUTISM SPECTRUM DISORDER (ASD) AND ATTENTION DEFICIT AND HYPERACTIVITY DISORDER (ADHD) CLASSIFICATION IN FEMALE COHORTS BASED ON RESTING-STATE FMRI IMAGES
AU - YI, Xinyao
AU - HUANG, Zhiguang
AU - HE, Yan
AU - Ming, Wai Kit
N1 - Full text of this publication does not contain sufficient affiliation information. With consent from the author(s) concerned, the Research Unit(s) information for this record is based on the existing academic department affiliation of the author(s)
PY - 2025/10
Y1 - 2025/10
N2 - Objectives: Emerging evidence suggests that Autism Spectrum Disorder (ASD) manifestations in females may present with subtler clinical features compared to male counterparts, while Attention Deficit and Hyperactivity Disorder (ADHD) in this demographic predominantly manifests through inattentive symptomatology rather than hyperactive-impulsive behaviors. Such phenotypic convergence contributes to substantial diagnostic challenges, rendering conventional assessment protocols particularly susceptible to misclassification in this population. We aim to propose a lightweight deep neural network, FNeuroNet, to accurately differentiate female ASD and ADHD individuals. Methods: Our methodology was benchmarked against prevalent ImageNet pre-trained models, including MobileNet, Xception, and ResNet18, for training on resting-state fMRI images. The FNeuroNet architecture was systematically engineered with optimized convolutional layers, strategically positioned maxpooling operations, and adaptive dropout mechanisms, specifically tailored to address the unique characteristics of resting-state fMRI data through careful consideration of its patterns. Results: The experimental results demonstrated that FNeuroNet achieved the best performance metrics on the testing set, attaining a classification accuracy of 0.8235, a precision of 0.7419, a recall of 0.9583, and an F1-score of 0.8364. Comparative analysis revealed that the proposed model significantly outperformed conventional pre-trained deep neural network architectures, establishing its superior discriminative capability in the target classification task. Conclusions: This system demonstrates significant potential in classifying ASD and ADHD between female. It augments the accuracy of clinical decision-making processes while improves patients' quality of life through timely intervention. Moreover, its multi-platform compatibility enables real-time processing of patient health records, thereby expediting diagnostic evaluations and therapeutic interventions to enhance healthcare delivery efficiency.
AB - Objectives: Emerging evidence suggests that Autism Spectrum Disorder (ASD) manifestations in females may present with subtler clinical features compared to male counterparts, while Attention Deficit and Hyperactivity Disorder (ADHD) in this demographic predominantly manifests through inattentive symptomatology rather than hyperactive-impulsive behaviors. Such phenotypic convergence contributes to substantial diagnostic challenges, rendering conventional assessment protocols particularly susceptible to misclassification in this population. We aim to propose a lightweight deep neural network, FNeuroNet, to accurately differentiate female ASD and ADHD individuals. Methods: Our methodology was benchmarked against prevalent ImageNet pre-trained models, including MobileNet, Xception, and ResNet18, for training on resting-state fMRI images. The FNeuroNet architecture was systematically engineered with optimized convolutional layers, strategically positioned maxpooling operations, and adaptive dropout mechanisms, specifically tailored to address the unique characteristics of resting-state fMRI data through careful consideration of its patterns. Results: The experimental results demonstrated that FNeuroNet achieved the best performance metrics on the testing set, attaining a classification accuracy of 0.8235, a precision of 0.7419, a recall of 0.9583, and an F1-score of 0.8364. Comparative analysis revealed that the proposed model significantly outperformed conventional pre-trained deep neural network architectures, establishing its superior discriminative capability in the target classification task. Conclusions: This system demonstrates significant potential in classifying ASD and ADHD between female. It augments the accuracy of clinical decision-making processes while improves patients' quality of life through timely intervention. Moreover, its multi-platform compatibility enables real-time processing of patient health records, thereby expediting diagnostic evaluations and therapeutic interventions to enhance healthcare delivery efficiency.
U2 - 10.1016/j.vhri.2025.101222
DO - 10.1016/j.vhri.2025.101222
M3 - RGC 32 - Refereed conference paper (with host publication)
T3 - Value in Health Regional issues
BT - ISPOR Real-World Evidence Summit 2025
ER -