Explainable Multimodal AI for Medical Imaging — Cardiovascular Imaging, Cancer Imaging and Cardiomyocyte Ageing. PhD researcher in Computer Science at the University of East London, developing explainable and multimodal AI for medical imaging. The work spans three imaging domains: cardiovascular imaging (cardiac MRI, 3D echocardiography, optical electrophysiology mapping), cancer imaging (dermoscopy and melanoma detection, breast cancer detection and prognosis, brain-tumour MRI, lung adenocarcinoma survival modelling), and cardiomyocyte ageing from live-cell microscopy. Across all three I combine segmentation and detection architectures, spatiotemporal modelling, multimodal data fusion, and clinically relevant validation to produce interpretable models. I am also an Hourly Paid Lecturer at UEL and an MSc Bioinformatics Dissertation Supervisor at Birkbeck, University of London, within the Institute of Structural and Molecular Biology (Birkbeck–UCL). Based in London, UK.

Google Scholar · ResearchGate · LinkedIn · Personal-Website · ORCID

1. Cardiovascular imaging. Transformer and hybrid CNN–RNN architectures for cardiac MRI and 3D echocardiography — segmentation, structural analysis, and label-efficient training from scribble annotations. Extends to optical mapping of cardiac electrophysiology and to fairness auditing of cardiovascular imaging models.

2. Cancer imaging. Vision–language and detection models for oncology imaging: melanoma and skin-lesion classification from dermoscopy (BCN20000, with cross-dataset validation on ISIC 2019), triple-negative breast cancer detection using YOLO-based models with hospital-cohort clinical validation, brain-tumour classification from MRI, and multivariable survival modelling in lung adenocarcinoma.

3. Cardiomyocyte ageing. Motion and nuclear phenotyping of cardiomyocytes from microscopy time-series, using optical flow and temporal transformers to detect functional decline before it is morphologically obvious.

4. Explainability and clinical validation. Post-hoc interpretability, algorithmic-bias auditing, and prospective validation of models against clinical endpoints — so that the output is defensible in a clinical setting, not only accurate on a benchmark.

Methodologically this means vision transformers, vision–language models, multimodal fusion (imaging + tabular + text), generative augmentation for small clinical cohorts, and survival/prognostic modelling.

Every modality below links to the repository where it is actually used, so the claim is checkable rather than asserted.

Project	What it is
triFuse-pytorch	TriFuse-SRNet — dynamic multi-expert fusion with structural recovery for scribble-supervised cardiac MRI segmentation. Reference PyTorch implementation, ACDC + MSCMRseg, with training, evaluation and statistical-comparison scripts.
medical-image-analysis	Streamlit dashboard for segmentation-quality assessment — Dice, IoU, Hausdorff distance, uncertainty and regional performance. Used to audit the outputs of TriFuse-SRNet.
Structured-Caption-Supervision-for-Domain-Adaptive-Vision-Language-Learning-	Cancer imaging. Melanoma and skin-lesion classification from dermoscopy using vision–language learning with structured clinical caption supervision — CLIP and BLIP fine-tuned on BCN20000, cross-dataset validation on ISIC 2019, with Grad-CAM explanations and ablations.
Breast cancer diagnosis and prognosis	Cancer imaging. Triple-negative breast cancer detection and prognostic analysis — YOLO-based detection, shrinkage operators, sequence networks and Kamada-Kawai graph analysis, with a clinical validation phase on hospital trial data.
Neuro-App	Cancer imaging. AI-driven 4D brain image processing and tumour classification from MRI, with GLCM/LBP texture metrics, sensitivity analysis under noise and blur, and 3D intensity visualisation.
AI-Models-for-Early-Cardiovascular-Diseases-Detection-	Early detection and mortality prediction in cardiovascular disease — the platform behind Diagnostics 14(12), 1308 (2024).
Advanced-Cardiac-Electrophysiology-Mapping-	Automated signal windowing and kriging-based spatial interpolation for optical mapping of cardiac electrophysiology — flecainide, low-flow ischaemia and cooling series.
Deep-Spatiotemporal-Modelling-of-Cardiomyocyte-Ageing-Dysfunction	Optical-flow–driven detection of cardiomyocyte ageing from microscopy video, with a Transformer over motion phenotypes. Presented at BSCR/BCS (Heart, 2025).

• Hybrid deep learning for computational precision in cardiac MRI segmentation: integrating autoencoders, CNNs and RNNs for enhanced structural analysis. Computers in Biology and Medicine 186, 109597 (2025).
• AI-driven thoracic X-ray diagnostics: transformative transfer learning for clinical validation in pulmonary radiography. Journal of Personalized Medicine 14(8), 856 (2024). DOI
• Mitigating algorithmic bias in AI-driven cardiovascular imaging for fairer diagnostics. Diagnostics 14(23), 2675 (2024). DOI
• Enhancing clinical validation for early cardiovascular disease prediction through simulation, AI and web technology. Diagnostics 14(12), 1308 (2024). DOI
• Hypertension control in resource-constrained settings: bridging socioeconomic gaps with predictive insights. IJC Cardiovascular Risk and Prevention (2025).
• Advanced transformer-based AI framework for early detection and prediction of cardiomyocyte ageing and injury using motion phenotyping. Heart 111 (Suppl 3), A269–A271 (2025).

Full list on Google Scholar. ORCID: 0009-0007-3503-6942.

Patent. 6426513 - Blockchain based health monitoring device, registered with the UK Intellectual Property Office.

Cardiovascular imaging and segmentation

• triFuse-pytorch — TriFuse-SRNet, scribble-supervised segmentation (ACDC, MSCMRseg)
• Cardiac-mri-scribble-segmentation — weakly supervised segmentation from scribble labels
• Generative-Augmentation-Using-VAE-GAN-and-Attention-U-Net — augmentation for small cardiac cohorts
• 3D-Heart-Imaging-apps — 3D cardiac imaging application
• Enhancing-Hypertrophic-Cardiomyopathy-Diagnosis-with-Deep-Learning-Driven-3D-Echocardiogram-Analysis
• Deep-Learning-Methods-for-Heart-Image-Analysis — review and benchmarking
• AI-Models-for-Early-Cardiovascular-Diseases-Detection- — code for Diagnostics 14(12), 1308 (2024)

Cancer imaging and oncology

• Structured-Caption-Supervision-for-Domain-Adaptive-Vision-Language-Learning- — melanoma and skin-lesion classification from dermoscopy; CLIP and BLIP with structured clinical captions, BCN20000 with external validation on ISIC 2019
• -Machine-Learning-Strategies-for-Breast-Cancer-Diagnosis-and-Prognosis — triple-negative breast cancer detection and prognosis; YOLO detection, shrinkage operators, sequence networks, hospital-cohort validation
• Neuro-App — 4D brain image processing and tumour classification from MRI, with texture metrics and robustness analysis
• Multivariable-Modelling-Reveals-EGFR-Copy-Number-as-an-Independent-Predictor-of-Survival-in-LUAD — survival modelling in lung adenocarcinoma
• Medical_research_sample_size_calculator- — sample-size calculator for oncology, cardiology and epidemiology study design

Cardiomyocyte ageing and cell-level phenotyping

• Deep-Spatiotemporal-Modelling-of-Cardiomyocyte-Ageing-Dysfunction
• Cardiomyocyte-cell-motion-analysis
• Cardiomyocyte-Ageing-Nucleus-Data-Analysis
• New-Apps-cardiomyocyte
• Cardiac-Cell-Development-Animation-Day-1-8-

Research software and clinical tools

• Advanced-Cardiac-Electrophysiology-Mapping- — automated signal windowing, parallel processing
• CardioMap-Pro-2.0-Advancing-Cardiac-Research-with-Next-Gen-Analytical-Software
• medical-image-analysis — segmentation-quality dashboard (Dice, IoU, Hausdorff, uncertainty)
• Heart-Failure-Prediction-Calculator
• biohybrid-heart-valve — design and performance dashboard for biohybrid valves

Explainability, multimodal and foundation models

• post-hoc-explainer — feature attribution and explanation of trained model inference
• High-Throughput-Multimodal-AI-Fusion-of-OCT- — hypergraph fusion of OCT and functional data for glaucoma detection
• BioNeuralNet-Leiden — graph-based multi-omics network analysis for biomarker discovery
• FinRAG — retrieval-augmented generation over documents

Clinical prediction in other domains

• TM-AHF-A-Text-Mining-a — text mining plus vital signs for BNP/NT-proBNP-validated acute heart failure risk stratification
• Transformative-Insights-in-Pulmonary-Radiography-AI-Enabled-Innovations — thoracic X-ray diagnostics
• Alzheimer-s-Disease-Prediction-Using-Handwriting-and-AI-Models-for-Cognitive-Assessment
• Early-Pregnancy-Detection-and-prevention-using-Advanced-machine-learning-Algorithms
• Survival-Analysis — survival modelling with synthetic-data fidelity evaluation

Fairness, health policy and population health

• -Cardiovascular-Health-Tackling-Algorithmic-Bias-in-ML-and-AI-Models
• Hypertension-Control-in-Resource-Constrained-Settings
• AI-Enabled-Study-of-Funding-Cuts-in-the-UK-Exploring-Regional-Mental-Health-Disparities
• Machine-Learning-and-Public-Health-Predictive-Analysis-of-UK-Traffic-Accident-Severity-and-I

Earlier applied data science (pre-PhD)

Kept for provenance; these predate the current research programme.

• Data-analytics-on-key-indicators-for-the-smart-city-s-urban-services-and-dashboards-for-leadership-a
• UK-s-Energy-Landscape-An-ESG-Perspective
• Machine-Learning-and-Sustainability-Metrics-Optimising-Risk-Assessment-and-Default-Prediction.
• Water-modelling — Bayesian uncertainty quantification for sediment-flux prediction
• A-Deep-Learning-based-Twitter-Sentiment-Analysis-for-Bitcoin-Market-Predictions-
• Developing-Trading-Strategies-in-Decentralized-Trading-Markets-Prediction-by-Using-AI-ML-And-Blockc
• Data-Analytics-for-Esports · Boost-Financial-Health-in-the-E-sports-Industry-Machine-Learning-Model-based
• Micro-Segmentation-Through-Hybrid-Clustering

• Hourly Paid Lecturer, University of East London — teaching and supervision of MSc and undergraduate projects
• Dissertation supervisor, MSc Bioinformatics, Birkbeck, University of London
• Editorial involvement, British Society of Cardiovascular Research
• Open to collaboration on cardiovascular imaging, cancer imaging, cardiomyocyte ageing and clinical validation of medical AI

Python · PyTorch · MONAI · scikit-learn · OpenCV · R · MATLAB · Streamlit · Docker · Git · LaTeX

Imaging: cardiac MRI, 3D echocardiography, dermoscopy, brain MRI, thoracic X-ray, retinal OCT, live-cell microscopy

abusufian.dev · LinkedIn · Google Scholar