Machine-learning-based feature selection to identify ADHD using white matter microstructure
Poster No:
340
Submission Type:
Abstract Submission
Authors:
Huey-Ling Chiang1, Chi-Shin Wu2, Susan Shur-Fen Gau3
Institutions:
1Far Eastern Memorial Hospital, New Taipei City, [Select a State], 2National Health Research Institutes, Taipei, Taiwan, 3National Taiwan University College of Medicine, Taipei, Taiwan
First Author:
Co-Author(s):
Introduction:
While brain imaging has been extensively used to investigate structural and functional alterations to provide objective measurements in attention-deficit/hyperactivity disorder (ADHD), the findings have exhibited considerable variability across studies with traditional univariate approaches. Although relatively few studies have used multi-modal image-based machine-learning approaches, including diffusion imaging, all of them reported that features of diffusion imaging provided specific importance in the model to improve discriminative power for ADHD diagnosis (Chaim-Avancini et al., 2017) Here, we aimed to identify white matter features collectively distinguishing individuals with ADHD from those without ADHD. We wanted to identify neuroimaging features associated with ADHD by examining the baseline, follow-up, and yearly change rate of white matter microstructure in a longitudinal dataset (Fuelscher et al., 2023). We hypothesize that imaging features from the white matter microstructure will enhance the accurate discrimination between individuals with a childhood ADHD diagnosis and typically developing controls (TDC).
Methods:
Fifty-one ADHD patients and 60 typically developing controls (TDC), underwent diffusion spectrum imaging at two time points. The generalized fractional anisotropy (GFA) was calculated for the microstructural properties of 45 white matter tracts. Machine-learning algorithms were utilized to classify ADHD and TDC. Three models were tested using machine-learning approaches. In the first model, we used baseline white matter features collected at Time 1 to classify the ADHD group from the TDC group. The second model included white matter features collected at both Time 1 and Time 2. The third model (main analysis) included a yearly change rate for each white matter tract. All analyses included age, sex, and image quality (measured by signal dropout count) as covariates. Correlation analyses were employed to depict the association between the yearly GFA values change rate and the neuropsychological performance changes for both the ADHD and TDC groups, respectively. These analyses involved selected features of ADHD in the classification model for distinguishing ADHD from TDC.
Results:
The random forest algorithm demonstrated the best performance for classification. Model 1 achieved an area-under-the-curve (AUC) of 0.67. Model 3, incorporating Time 2 variables and yearly change rates, improved the performance (AUC=0.73). In addition to identifying several white matter features at two time points, we found that the yearly change rates in the superior longitudinal fasciculus, frontal aslant tract, stria terminalis, inferior fronto-occipital fasciculus, thalamic and striatal tracts, and other tracts involving sensorimotor regions are important features of ADHD (Figure 1). Correlation analyses indicated that higher yearly increasing GFA rates in certain tracts were associated with greater improvement in visual attention, spatial short-term memory, and spatial working memory after FDR corrections (Table 1).
Conclusions:
Using longitudinal DSI data of white matter microstructure, this machine-learning-based analysis achieves moderate discrimination power in classifying individuals with and without childhood ADHD diagnosis. The properties of white matter white matter microstructure and its developmental change rate, which reflect deviations from typical development trajectories, serve as important biomarkers for ADHD.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 1
Lifespan Development:
Early life, Adolescence, Aging 2
Keywords:
Attention Deficit Disorder
WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC
Other - machine-learning
1|2Indicates the priority used for review
·Figure 1. The white matter features to differentiate individuals with childhood diagnosis of ADHD from those with typical development
·Table 1. Correlation between the yearly change rate of the GFA and the development of neuropsychological function by group
Provide references using author date format
Fuelscher, I., Hyde, C., Thomson, P., Vijayakumar, N., Sciberras, E., Efron, D., Anderson, V., Hazell, P., Silk, T.J., 2023. Longitudinal Trajectories of White Matter Development in Attention-Deficit/Hyperactivity Disorder. Biol Psychiatry Cogn Neurosci Neuroimaging: S2451-9022(23)00071-X.