Deep Transfer Learning for Cerebral Cortex Using Area-Preserving Geometry Mapping.

Within the light of the DSM-5, the current study examined (1) how many and which children with a DSM-IV classification of autism spectrum disorder (ASD) fulfill the DSM-5 symptom-criteria, and (2) whether children who did and did not meet DSM-5 symptom-criteria and children with social anxiety disorder (SAD) can be differentiated from each other based on ASD symptomatology. In total, 90 referred children with a DSM-IV classification of high-functioning ASD, and 21 referred children with SAD participated (age range 7–17 years). ASD-symptoms were examined with the Autism Diagnostic Interview-Revised and the Children’s Social Behavior Questionnaire. It was found that 30 % of the ASD sample did not meet DSM-5 symptom-criteria for ASD, mainly because they failed to meet the DSM-5 criteria of the repetitive domain. Children with ASD who did and did not meet DSM-5 symptom-criteria differed on the repetitive domain, while children with ASD (according to DSM-IV and DSM-5 symptom criteria) had higher scores on the social-communication domain than children with SAD. Findings suggest a continuum of ASD-symptoms in the DSM-5 for children with SAD, social communication disorder and ASD. More research is needed to examine how these three disorders differ with respect to their etiology, neuropsychological profiles and clinical characteristics.

Autism Spectrum Disorder (ASD) is a lifelong neurodevelopmental condition characterized by challenges in social communication and repetitive behaviors, making early diagnosis essential for effective intervention. This preliminary study examines the feasibility of classifying ASD using 2D slices extracted from 3D brain MRI volumes, focusing on the comparative diagnostic value of axial, coronal, and sagittal planes within a machine learning framework. The goal is to identify the most informative anatomical plane to guide future neuroimaging protocols and improve automated ASD detection. We analyzed 20,992 slices derived from 82 MRI scans (41 ASD and 41 non-ASD) from the Autism Brain Imaging Data Exchange (ABIDE), each scan averaging 256 × 256 pixels across 256 slices. Preprocessing included skull stripping and conversion to NIFTI-GZ format. Texture features were extracted using the Gray Level Co-occurrence Matrix (GLCM), and classification was performed using the K-Nearest Neighbors (KNN) algorithm with varying K values (3, 5, and 7). The best performance was achieved with K = 3 and GLCM parameters at distance d = 0 and angle 45°, yielding an accuracy of 83.90%, precision of 79%, recall of 78%, and F1-score of 78%. These findings provide initial evidence that 2D slice-based texture analysis can support early ASD classification and offer practical insights for optimizing neuroimaging-based diagnostic tools. Future work will explore deep learning approaches and multimodal integration to enhance robustness and clinical applicability.

SummaryBackgroundDiagnosis of autism spectrum disorder (ASD) significantly improves quality of life, yet current diagnostic practices rely on subjective behavioural assessments. While machine learning models using functional Magnetic Resonance Imaging (fMRI) show promise for objective diagnosis, existing approaches have critical limitations. High-accuracy models often neglect interpretability, increasing clinical distrust, while interpretable approaches frequently suffer from low accuracy and lack neuroscientific validation of identified biomarkers. Furthermore, no studies have systematically benchmarked which interpretability methods are most effective for fMRI data, hindering the translation of Artificial intelligence (AI) findings into clinical practice.MethodsIn this observational study, we used the Autism Brain Imaging Data Exchange I (ABIDE I) dataset comprising 884 participants (408 with ASD; 476 controls with typical development) aged 7–64 years from five countries (Belgium, Germany, Ireland, the Netherlands, and the USA) across 17 international sites after applying mean framewise displacement (FD) filtering (>0.2 mm). Data were collected at the respective sites prior to the release of the ABIDE I dataset in August 2012. We described an explainable deep learning (DL) pipeline using a Stacked Sparse Autoencoder (SSAE) with a softmax classifier, trained on functional connectivity data. We systematically benchmarked seven interpretability methods using the Remove And Retrain (ROAR) technique and cross-validated our findings across three preprocessing pipelines. Critically, we validated identified biomarkers against independent neuroscientific literature from genetic, neuroanatomical, and functional studies.FindingsFiltering head movement (FD > 0.2 mm) increased classification accuracy from 91% to 98.2% (F1-score: 0.97), achieving state-of-the-art performance. ROAR analysis revealed gradient-based methods, particularly Integrated Gradients, as most reliable for fMRI interpretation. The model consistently identified visual processing regions (calcarine sulcus, cuneus) as critical for ASD classification. These findings aligned with independent genetic studies and neuroimaging research, confirming our model captured genuine neurobiological ASD markers rather than overfitting to dataset artifacts.InterpretationWe addressed three key research gaps by developing a highly accurate, interpretable ASD classification model, establishing a benchmarking framework for interpretability methods in the fMRI modality, and validating our identified biomarkers against established neuroscience literature. The consistent importance of visual processing regions suggests a fundamental biomarker potentially present across the ASD spectrum, offering new directions for diagnostic biomarker research, though translation to individual-level clinical applications requires further development.FundingThis work was supported by the Engineering and Physical Sciences Research Council, UK [EP/W524554/1].

Deep learning has attracted increasing attention in brain imaging, but many neuroimaging data samples are small and fail to meet the training data requirements to optimize performance. In this study, we propose a deep transfer learning network based on Optimal Mass Transport (OMTNet) for 3D brain image classification using MRI scans from the UK Biobank. The major contributions of the OMTNet method include: a way to map 3D surface-based vertex-wise brain shape metrics, including cortical thickness, surface area, curvature, sulcal depth, and subcortical radial distance and surface Jacobian determinant metrics, onto 2D planar images for each MRI scan based on area-preserving mapping. Such that some popular 2D convolution neural networks pretrained on the ImageNet database, such as ResNet152 and DenseNet201, can be used for transfer learning of brain shape metrics. We used a score-fusion strategy to fuse all shape metrics and generate an ensemble classification. We tested the approach in a classification task conducted on 26k participants from the UK Biobank, using body mass index (BMI) thresholds as classification labels (normal vs. obese BMI). Ensemble classification accuracies of 72.8 ± 1.2% and 73.9 ± 2.3% were obtained for ResNet152 and DenseNet201 networks that used transfer learning, with 5.4–12.3% and 6.1–13.0% improvements relative to classifications based on single shape metrics, respectively. Transfer learning always outperformed direct learning and conventional linear support vector machines with 3.4–8.7% and 4.9–6.0% improvements in ensemble classification accuracies, respectively. Our proposed OMTNet method may offer a powerful transfer learning framework that can be extended to other vertex-wise brain structural/functional imaging measures.

The use of functional magnetic resonance imaging (fMRI) data in machine learning (ML)-based classification of autism spectrum disorder (ASD) has been a topic of increasing research interest in past years due to the noninvasiveness of the fMRI technique and its potential for providing valuable biomarkers. However, there are still controversies surrounding some fMRI data preprocessing steps, such as bandpass filtering and global signal regression (GSR). It still needs to be determined whether or how these preprocessing steps impact the classification accuracy of ML algorithms. This paper uses fMRI signals from the ABIDE-I dataset to train a long short-term memory (LSTM) network to classify subjects into ASD or healthy controls (HC). We considered 18 preprocessing pipelines comprising all combinations of with and without filtering, with and without global signal regression, and three different segment lengths of 1 min, 2 min, and 3 min. The best model was obtained when using a segment length of 2 min. Our results suggest that not filtering produces significantly higher classification accuracies than filtering, whereas there were no significant differences in classification accuracies when removing or not the global signal.

Autism spectrum disorder (ASD) is a neurodevelopmental disorder characterized by a lack of social communication and social interaction. Autism is a mental disorder investigated by social and computational intelligence scientists utilizing advanced technologies such as machine learning models to enhance clinicians’ ability to provide robust diagnosis and prognosis of autism. However, with dynamic changes in autism behaviour patterns, these models’ quality and accuracy have become a great challenge for clinical practitioners. We applied a deep neural network learning on a large brain image dataset obtained from ABIDE (autism brain imaging data exchange) to provide an efficient diagnosis of ASD, especially for children. Our deep learning model combines unsupervised neural network learning, an autoencoder, and supervised deep learning using convolutional neural networks. Our proposed algorithm outperforms individual-based classifiers measured by various validations and assessment measures. Experimental results indicate that the autoencoder combined with the convolution neural networks provides the best performance by achieving 84.05% accuracy and Area under the Curve (AUC) value of 0.78.

Ultra-high dimensional variable selection has become increasingly important in analysis of neuroimaging data. For example, in the Autism Brain Imaging Data Exchange (ABIDE) study, neuroscientists are interested in identifying important biomarkers for early detection of the autism spectrum disorder (ASD) using high resolution brain images that include hundreds of thousands voxels. However, most existing methods are not feasible for solving this problem due to their extensive computational costs. In this work, we propose a novel multiresolution variable selection procedure under a Bayesian probit regression framework. It recursively uses posterior samples for coarser-scale variable selection to guide the posterior inference on finer-scale variable selection, leading to very efficient Markov chain Monte Carlo (MCMC) algorithms. The proposed algorithms are computationally feasible for ultra-high dimensional data. Also, our model incorporates two levels of structural information into variable selection using Ising priors: the spatial dependence between voxels and the functional connectivity between anatomical brain regions. Applied to the resting state functional magnetic resonance imaging (R-fMRI) data in the ABIDE study, our methods identify voxel-level imaging biomarkers highly predictive of the ASD, which are biologically meaningful and interpretable. Extensive simulations also show that our methods achieve better performance in variable selection compared to existing methods.

Autism spectrum disorder (ASD) encompasses a group of neurodevelopmental conditions diagnosed solely on the basis of behavioral assessments that reveal social deficits. Progress has been made in understanding its genetic underpinnings, but most ASD-associated genetic variants, which include copy number variants (CNVs) and mutations in ASD-risk genes, account for no more than 1 % of ASD cases. This high level of genetic heterogeneity leads to challenges obtaining and interpreting genetic testing in clinical settings. The traditional definition of syndromic ASD is a disorder with a clinically defined pattern of somatic abnormalities and a neurobehavioral phenotype that may include ASD. Most have a known genetic cause. Examples include fragile X syndrome and tuberous sclerosis complex. We propose dividing syndromic autism into the following two groups: (i) ASD that occurs in the context of a clinically defined syndrome-recognizing these disorders depends on the familiarity of the clinician with the features of the syndrome, and the diagnosis is typically confirmed by targeted genetic testing (eg, mutation screening of FMR1); (ii) ASD that occurs as a feature of a molecularly defined syndrome-for this group of patients, ASD-associated variants are identified by genome-wide testing that is not hypothesis driven (eg, microarray, whole exome sequencing). These ASD groups cannot be easily clinically defined because patients with a given variant have variable somatic abnormalities (dysmorphism and birth defects). In this article, we review common diagnoses from the above categories and suggest a testing strategy for patients, guided by determining whether the individual has essential or complex ASD; patients in the latter group have multiple morphologic anomalies on physical examination. Finally, we recommend that the syndromic versus nonsyndromic designation ultimately be replaced by classification of ASD according to its genetic etiology, which will inform about the associated spectrum and penetrance of neurobehavioral and somatic manifestations.

Autism spectrum disorder (ASD) affects social interaction, communication, and behavior. Early diagnosis is important as it enables timely intervention that can significantly improve long-term outcomes, but current diagnostic, which rely heavily on behavioral observations and clinical interviews, are often subjective and time-consuming. This study introduces an AI-based approach that uses T1-weighted structural MRI (sMRI) scans, network analysis, and vision transformers to automatically diagnose ASD. sMRI data from 79 ASD patients and 105 healthy controls were obtained from the Autism Brain Imaging Data Exchange (ABIDE) database. Complex network analysis (CNA) features and ViT (Vision Transformer) features were developed for predicting ASD. Five models were developed for each type of features: logistic regression, support vector machine (SVM), gradient boosting (GB), K-nearest neighbors (KNN), and neural network (NN). 25 models were further developed by federating the two sets of 5 models. Model performance was evaluated using accuracy, area under the receiver operating characteristic curve (AUC-ROC), sensitivity, and specificity via fivefold cross-validation. The federate model CNA(KNN)-ViT(NN) achieved highest performance, with accuracy 0.951 ± 0.067, AUC-ROC 0.980 ± 0.020, sensitivity 0.963 ± 0.050, and specificity 0.943 ± 0.047. The performance of the ViT-based models exceeds that of the complex network-based models on 80% of the performance metrics. By federating CNA models, the ViT models can achieve better performance. This study demonstrates the feasibility of using CNA and ViT models for the automated diagnosis of ASD. The proposed CNA(KNN)-ViT(NN) model achieved better accuracy in ASD classification based solely on T1 sMRI images. The proposed method's reliance on widely available T1 sMRI scans highlights its potential for integration into routine clinical examinations, facilitating more efficient and accessible ASD screening.

Autism Spectrum Disorder (ASD) was characterized by delay in social interactions development, repetitive behaviors and narrow interest, which usually diagnosed with standard diagnostic tools such as Autism Diagnostic Observation Schedule (ADOS) and Autism Diagnostic Interview-Revised (ADIR-R). Previous work has implemented machine-learning methods for the classification of ASD, however they used different types of dataset such as brain images for MRI and EEG, risk genes in genetic profiles and behavior evaluation based on ADOS and ADI-R. Here a trial on using Autism Spectrum Questions (AQ) to build models that have higher potential to classify ASD was developed. In this research, Chi-square and Least Absolute Shrinkage and Selection Operator (LASSO) have been selected as feature selection methods to select the most important features for 3 supervised machine learning algorithms, which are Random Forest, Logistic Regression and K-Nearest Neighbors with K-fold cross validation. The performance was evaluated in which results Logistic Regression scored the highest accuracy with 97.541% using model with 13 selected features based on Chi-square selection method.

The presence of multiple dysmorphic features in some children with autism spectrum disorder (ASD) might identify distinct ASD phenotypes and serve as potential markers for understanding causes and prognoses. To evaluate dysmorphology in ASD, children aged 3-6years with ASD and non-ASD population controls (POP) from the Study to Explore Early Development were evaluated using a novel, systematic dysmorphology review approach. Separate analyses were conducted for non-Hispanic White, non-Hispanic Black, and Hispanic children. In each racial/ethnic group, ~ 17% of ASD cases were Dysmorphic compared with ~ 5% of POP controls. The ASD-POP differential was not explained by known genetic disorders or birth defects. In future epidemiologic studies, subgrouping ASD cases as Dysmorphic vs. Non-dysmorphic might help delineate risk factors for ASD.

Autism spectrum disorder (ASD) encompasses a wide range of complex neurodevelopment conditions, posing significant challenges to early diagnosis. In fact, other than early diagnosis (i.e., before normal development) it is extremely challenging to adequately classify these subjects as there are no behavioral markers that are unique to ASD vs. non-ASD subjects. But the early detection is very important in helping with some of the developmental trajectories as the timely intervention which they are getting and the access which the kids and families have to specially designed therapies and support systems will be better. Neuroimaging in autism: a comprehensive review from a machine learning perspective. Autism Research- We investigate the effectiveness of CNN models and a hybrid CNN-RNN architecture in classifying ASD based on the neuroimaging data with an aim of improving feature extraction from the neuroimaging data using the Automated Anatomical Labeling (AAL) map. A key challenge in this task is how to build representational feature maps that can elucidate the intricate patterns across brain regions linked with ASD. Various kinds of models are tested VGG16 2D CNN, VGG16 3D CNN, EfficientNetV2, Inception v3, ResNet50 2D CNN and Hybrid model (CNN-RNN). While going for traditional CNN models gave at least 60% to 69% accuracies, the accuracy from hybrid CNN-RNN model, which combines spatial and temporal features, far outperformed other models with an accuracy of 100%. Thus, hybrid deep learning architectures have a substantial role in the classification of ASD. An involvement of AAL mapping and deep learning methods in this study proposed that the semi-automated approach could be implemented in forensic precision for the early diagnosis of ASD. These findings guide future neuroimaging-based diagnostic research and support the development of more sophisticated hybrid models for classifying autism spectrum disorder.

Learning graph-based relationship of dual-modal features towards subject adaptive ASD assessment

Neurodevelopmental conditions, such as Autism Spectrum Disorder (ASD) and Attention Deficit Hyperactivity Disorder (ADHD), present unique challenges due to overlapping symptoms, making an accurate diagnosis and targeted intervention difficult. Our study employs advanced machine learning techniques to analyze functional magnetic resonance imaging (fMRI) data from individuals with ASD, ADHD, and typically developed (TD) controls, totaling 120 subjects in the study. Leveraging multiclass classification (ML) algorithms, we achieve superior accuracy in distinguishing between ASD, ADHD, and TD groups, surpassing existing benchmarks with an area under the ROC curve near 98%. Our analysis reveals distinct neural signatures associated with ASD and ADHD: individuals with ADHD exhibit altered connectivity patterns of regions involved in attention and impulse control, whereas those with ASD show disruptions in brain regions critical for social and cognitive functions. The observed connectivity patterns, on which the ML classification rests, agree with established diagnostic approaches based on clinical symptoms. Furthermore, complex network analyses highlight differences in brain network integration and segregation among the three groups. Our findings pave the way for refined, ML-enhanced diagnostics in accordance with established practices, offering a promising avenue for developing trustworthy clinical decision-support systems.

The classification of autism spectrum disorder by machine learning methods on multiple datasets for four age groups