Single-participant structural similarity matrices lead to greater accuracy in classification of participants than function in autism in MRI

Abstract

Background Autism has previously been characterized by both structural and functional differences in brain connectivity. However, while the literature on single-subject derivations of functional connectivity is extensively developed, similar methods of structural connectivity or similarity derivation from T1 MRI are less studied. Methods We introduce a technique of deriving symmetric similarity matrices from regional histograms of grey matter volumes estimated from T1-weighted MRIs. We then validated the technique by inputting the similarity matrices into a convolutional neural network (CNN) to classify between participants with autism and age-, motion-, and intracranial-volume-matched controls from six different databases (29,288 total connectomes, mean age = 30.72, range 0.42–78.00, including 1555 subjects with autism). We compared this method to similar classifications of the same participants using fMRI connectivity matrices as well as univariate estimates of grey matter volumes. We further applied graph-theoretical metrics on output class activation maps to identify areas of the matrices that the CNN preferentially used to make the classification, focusing particularly on hubs. Limitations While this study used a large sample size, the majority of data was from a young age group; furthermore, to make a viable machine learning study, we treated autism, a highly heterogeneous condition, as a binary label. Thus, these results are not necessarily generalizable to all subtypes and age groups in autism. Results Our models gave AUROCs of 0.7298 (69.71% accuracy) when classifying by only structural similarity, 0.6964 (67.72% accuracy) when classifying by only functional connectivity, and 0.7037 (66.43% accuracy) when classifying by univariate grey matter volumes. Combining structural similarity and functional connectivity gave an AUROC of 0.7354 (69.40% accuracy). Analysis of classification performance across age revealed the greatest accuracy in adolescents, in which most data were present. Graph analysis of class activation maps revealed no distinguishable network patterns for functional inputs, but did reveal localized differences between groups in bilateral Heschl’s gyrus and upper vermis for structural similarity. Conclusion This study provides a simple means of feature extraction for inputting large numbers of structural MRIs into machine learning models. Our methods revealed a unique emphasis of the deep learning model on the structure of the bilateral Heschl’s gyrus when characterizing autism.

MIND Networks: Robust Estimation of Structural Similarity from Brain MRI

“…Traditionally, KL divergence between empirical distributions is calculated by a two-step approach: i) non-parametric estimation of the probability density functions (PDFs) of the observed data; and ii) computing the divergence using the approximated PDFs. However, the initial density estimation step of this approach is sensitive to many choices of parameters (Leming et al, 2021; Homan et al, 2019; Wang et al, 2016; Perez-Cruz, 2008). Extended to multiple dimensions, density estimation becomes especially problematic; for multivariate data with as few as three dimensions, standard non-parametric density estimators provide very poor results (Wang and Scott, 2019).…”

“…Despite the promise of MSNs, one key limitation is that they reduce the rich, vertex-level data from MRI-based cortical surface reconstructions to single summary statistics for each feature per region. While other work has explored structural similarity measured directly from vertex-level data, these methods were limited to the use of a single structural feature such as cortical thickness (Homan et al, 2019) or grey matter volume (Leming et al, 2021; Kong et al, 2015).…”

Machine learning for autism spectrum disorder diagnosis using structural magnetic resonance imaging: Promising but challenging

“…However, in Wang et al (2021) , the maps of self-attention coefficients in the first and second layers are similar, indicating a consistent diagnosis. In Leming et al (2021) , the authors verified their suggested approach for classifying individuals with ASD and age-, motion-, and intracranial-volume-matched HCs by feeding a CNN the symmetric similarity matrix from regional histograms of estimated GM volumes. They also used graph-theoretic metrics on output CAMs to determine CNN’s favorite categorization regions, focusing on hubs.…”

Automatic autism spectrum disorder detection using artificial intelligence methods with MRI neuroimaging: A review

“…Computer-aided design systems based on ML algorithms are highly time-consuming and complex to design. However, if the frontiersin.org Moridian et al 10.3389/fnmol.2022.999605 appropriate algorithms are selected, it can accurately diagnose ASD (Iglesias et al, 2017;Khosla et al, 2019;Hiremath et al, 2020;Leming et al, 2020Leming et al, , 2021. DL methods automatically perform the steps from feature extraction to classification.…”