Research Projects

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition that currently lacks reliable biological markers for early diagnosis, particularly during infancy when interventions may be most effective. Most existing diagnostic systems rely on behavioral assessments administered after the age of two, often delaying the opportunity for early therapeutic intervention. Recent advances in neuroimaging and machine learning have opened the door to objective, imaging-based approaches to detect ASD-related abnormalities earlier in life. However, current imaging-based ASD studies remain limited in scope: they typically use a single modality, focus on children rather than infants, ignore developmental trajectories, and lack clinical interpretability or readiness. This project proposes a transformative approach to early ASD detection in infants (ages 6–24 months) through the development of a multimodal, interpretable deep learning framework using longitudinal neuroimaging data. By integrating structural MRI (sMRI), diffusion MRI (dMRI), and resting-state functional MRI (rs-fMRI), we aim to build models that not only classify ASD risk but also explain the neurobiological underpinnings of that risk across time. The rationale is rooted in both prior findings of atypical brain growth in ASD infants and the clinical demand for trustworthy, generalizable, and early-deployable diagnostic systems. Main Experiments 1. Preprocessing and feature extraction from sMRI, dMRI, and rs-fMRI using established neuroimaging tools (FreeSurfer, FSL, fMRIPrep). 2. Development of baseline deep learning models (CNNs, GCNs) for each modality, followed by a multimodal fusion network. 3. Temporal trajectory modeling using recurrent or transformer-based architectures. 4. Explainability integration and hybridization with neurobiological features (e.g., FA, cortical volumes). 5. Prototype development of a clinician-facing CADx tool and evaluation using usability testing. Significance to the Field If successful, this project will establish one of the first clinically relevant, interpretable, multimodal AI systems for early ASD detection in infants. It will bridge the gap between behavioral and biological diagnosis, offering a scientific tool for discovering developmental biomarkers and a practical solution for early intervention. The proposed methodology could set a precedent for future neurodevelopmental disorder diagnosis and catalyze the development of generalized, AI-based screening platforms. Furthermore, it lays the groundwork for foundation models that span multiple brain disorders, and for future deployment in clinical and public health settings, including regions like India.