Research

My research bridges applied mathematics and clinical oncology to build quantitative, actionable tools for precision oncology — from patient-specific tumour growth simulation to AI-driven pathology and imaging biomarkers. Over the past twenty years I have worked on three complementary methodological pillars, with a growing emphasis on the applications of artificial intelligence and mechanistic modelling to cancer diagnosis, prognosis, and treatment monitoring.

Mechanistic modelling of tumour growth

I develop reaction-diffusion PDE models (advection-reaction or reaction-diffusion) that capture the spatio-temporal dynamics of tumour proliferation, invasion, vascularisation, and response to cytotoxic or targeted therapies. A key challenge is patient-specific calibration from longitudinal CT or MRI data, using mixed-effects statistical frameworks or Bayesian data assimilation. Applications span soft-tissue sarcomas, renal tumours, gliomas, and meningiomas.

Deep learning for histological and radiological analysis

Modern deep learning architectures — CNNs, Vision Transformers, graph networks, and multiple-instance learning (MIL) frameworks — allow extraction of high-dimensional predictive features from whole-slide histopathological images and multi-parametric MRI/CT acquisitions. I design and validate end-to-end pipelines for automated tumour grading, subtype classification, and treatment-response prediction, with systematic uncertainty quantification and explainability (GradCAM, attention maps) and prospective multi-centric validation.

Radiomics and non-invasive tumour characterisation

Radiomics extracts quantitative texture, shape, and intensity features from routine medical images that reflect underlying tumour biology invisible to the naked eye. I build reproducible, open-source radiomics pipelines and validate them for early prediction of chemotherapy response, histological subtype prediction, and longitudinal treatment monitoring — across sarcomas, renal tumours, gliomas, hepatocellular carcinoma, and haematological malignancies.

Recent highlights

The most recent work focuses on multimodal AI and its integration with mechanistic knowledge for oncology:

• Multimodal integration for glioma recurrence prediction — combining transcriptomics, proteomics, and radiomics to predict recurrence in IDH-mutant gliomas (International Journal of Cancer, 2025)
• Deep learning prognosis in gynecologic smooth muscle tumours — multicenter study showing that deep learning on histological slides accurately predicts prognosis in tumours of uncertain malignant potential (Laboratory Investigation, 2025)
• Digital pathology for soft-tissue sarcoma prognosis — prognostic models from whole-slide images of tumour and surgical margin areas (Scientific Reports, 2025)
• Tumour growth models for meningioma — systematic comparison of four mathematical growth models for the natural history of meningiomas (EBioMedicine, 2023)