Research

My research brings together medical imaging AI, quantitative biomarkers and clinical evaluation. I work across structural and diffusion MRI and thoracic CT, connecting methodological development with questions in diagnosis, disease characterisation and clinical translation.

Neuroimaging & neurological disease

Epilepsy: lesion detection and segmentation

My research develops MRI and machine learning methods to identify and delineate subtle brain abnormalities associated with focal epilepsy, particularly focal cortical dysplasia. This includes unsupervised lesion detection using ultra-high-field 7T MRI and multimodal segmentation using T1-weighted and FLAIR images. I focus on detecting small lesions, understanding model failures and evaluating performance across imaging settings.

Predicting seizure recurrence after a first seizure

A first unprovoked seizure raises an important clinical question: will another seizure occur? My research investigates whether structural MRI, combined with clinical information, can help estimate this risk. Through NeuroMorphix, I study brain asymmetry and other imaging features, with evaluation in an independent patient cohort to assess how well predictions generalise.

Alzheimer’s disease: structural MRI biomarkers

My research investigates how structural MRI can characterise brain changes across normal cognition, mild cognitive impairment and Alzheimer’s disease. Using cortical and subcortical measurements, I examine regional anatomical differences and networks that describe similarities between brain regions. This work combines machine learning with harmonisation across imaging sites to evaluate potential biomarkers for distinguishing clinical groups.

Friedreich’s ataxia: diffusion MRI and cerebellar pathways

My research uses diffusion MRI and deep learning to study the cerebellar pathways affected by Friedreich’s ataxia. I develop automated methods to segment the cerebellar peduncles and extract quantitative measurements of white matter microstructure. The work evaluates segmentation accuracy and consistency across imaging sites, with the aim of supporting reproducible imaging biomarkers of neurological disease.

Occupational lung disease & thoracic imaging

Imaging in dust-exposed workers

At I-MED Radiology Network, I lead research delivery and analysis in occupational lung imaging. ExactDust examines agreement and diagnostic performance between chest radiography and HRCT. Related work focuses on quantitative HRCT assessment of structural lung abnormalities.

Connecting lung structure and function

My ACARP-supported research evaluates CT-derived measures of lung function against spirometry and other available pulmonary function measures in coal mine dust lung disease.

Clinical AI evaluation & translation

Across these projects, I focus on validation in clinical cohorts, multi-site variability, harmonisation, error analysis and transparent reporting. My experience spans clinical data curation, imaging quality control, segmentation, statistical analysis and reproducible machine learning pipelines.

Methods: MRI and CT analysis · Deep learning · Quantitative biomarkers · Multi-site harmonisation · Diagnostic performance evaluation

For research collaborations: soumen.ghosh@uq.edu.au