Research
Research Areas
Segmentation, diffusion models, benchmarks, and longitudinal lesion analysis.
02Brain MRI progression, atrophy mapping, and brain-age estimation beyond MS.
03Chest CT and X-ray models for detection, assessment, and infection estimation.
04Additional single-paper lines spanning orthopedics, oncology, and radiogenomics.
Multiple Sclerosis
Medical Image Segmentation
This work presents a volumetric deep architecture for multiple sclerosis lesion segmentation that combines squeeze-and-attention, deep supervision, and atrous spatial pyramid pooling to better capture 3D context, hierarchical features, and heterogeneous lesion patterns across multimodal MRI scans.
View PDF
Dataset & Benchmark
The paper introduces a publicly accessible MRI dataset for multiple sclerosis lesion segmentation with expert annotations, multimodal acquisitions, and baseline evaluations, providing a reliable benchmark for comparing automated segmentation methods against human-labelled references.
View PDF
Competition Report
This report summarizes the ICPR competition on automatic multiple sclerosis lesion segmentation, describing a heterogeneous baseline and follow-up MRI benchmark designed to promote robust lesion detection methods without user interaction across unseen cohorts and timepoints.
View PDF
Diffusion-Based Segmentation
This study compares diffusion-based segmentation strategies for multiple sclerosis lesions, focusing on how these models behave across multimodal MRI settings and how effectively they recover subtle lesion patterns in comparison with reference segmentation pipelines.
View PDF
Brain Parcellation
The paper studies how brain parcellation affects multiple sclerosis lesion segmentation, showing that parcellated regions can improve robustness and stability, particularly for small anomalies, even when mean segmentation gains remain limited.
View PDF
Diffusion Models
This work investigates denoising diffusion models for pixel-wise multiple sclerosis lesion segmentation, showing improved sensitivity to subtle abnormalities and competitive performance with state-of-the-art approaches across different MRI modalities.
View PDF
Attention Mechanisms
The paper proposes an augmented U-Net with convolutional LSTM and attention modules for automatic segmentation and quantification of multiple sclerosis lesions, reporting strong Dice performance and good robustness on unseen cases from a dedicated dataset.
View PDFCOVID-19
COVID-19 CT Analysis
This contribution presents an attention-based convolutional network for COVID-19 detection from chest CT scans, with the goal of supporting earlier diagnosis and management through more accurate and automatic analysis of pulmonary imaging patterns.
View PDF
Data Augmentation
The paper applies Inception-v3 together with mixup augmentation to estimate COVID-19 infection percentage from chest CT slices, introducing mixup on radiological lung images and obtaining promising challenge results despite strict competition constraints.
View PDF
X-Ray COVID-19 Assessment
This work investigates a deep non-local embedded system for automatic COVID-19 assessment from chest X-ray imaging, coupling 3D deep learning and self-augmentation to support accurate classification with widely available and low-cost radiological acquisitions.
View PDFNeuroimaging
Neurodegenerative Progression
The paper introduces a 3D diffusion-based model to predict future brain changes from MRI scans, addressing limitations of existing methods such as poor handling of time intervals, reliance on simple interpolation, and lack of full 3D context. The model leverages a pre-trained brain-age estimator to guide temporal progression and includes bidirectional temporal regularization to improve temporal consistency. It is trained on OASIS-3 and validated on NACC, showing good generalization performance.
View PDF
Neurodegenerative Progression
The proposed diffusion model predicts future brain MRIs by learning structural intensity changes over time and refining generation with a brain-age estimator, improving similarity and regional error metrics for modelling neurodegenerative progression.
View PDF
Brain Change Estimation
This contribution targets rapid estimation of structural brain changes through a single-step U-Net pipeline, aiming to make atrophy mapping faster and more practical for longitudinal neuroimaging analyses while preserving clinically relevant sensitivity.
View PDF
Brain Age Estimation
SynthBA introduces a robust deep-learning framework for brain age prediction that uses domain randomization to handle different scanners, sequences, and resolutions, enabling strong generalization across heterogeneous clinical MRI settings without retraining.
View PDFOthers
Cancer Outcome Prediction
This study proposes a dopamine-inspired neuro-modulated radiomic pipeline for predicting immunotherapy response from chest-abdomen CT imaging in metastatic cancer, achieving high accuracy in the reported clinical evaluation.
View PDF
Orthopedic Imaging
This study introduces a convolutional neural network for early detection of hip periprosthetic joint infections from CT imaging, addressing a task not previously explored and achieving high accuracy and F-score on a cohort of more than one hundred patients.
View PDF
Cancer Outcome Prediction
This study proposes a non-invasive 3D deep pipeline with attention and reinforcement learning to predict immunotherapy response from chest-abdomen CT imaging in metastatic bladder cancer, reaching accuracy close to 93% in the reported clinical evaluation.
View PDF
Immunotherapy Assessment
The authors develop a 3D densely connected deep network for selecting and augmenting spatio-temporal features from chest-abdomen CT images, improving prediction of immunotherapy outcomes for aggressive metastatic urothelial carcinoma.
View PDF
Radiogenomics
This paper proposes a deep pipeline for lesion segmentation and mutation-aware classification in lung CT scans, combining 3D Tiramisu segmentation, multi-scale classification, and adversarial training to improve interpretability while achieving strong segmentation and classification accuracy.
View PDF
Non-Invasive Biomarkers
The article presents a non-invasive deep-learning biomarker that integrates CT imaging and patient-related data to predict immunotherapy response in metastatic urothelial carcinoma, showing promising accuracy for early treatment outcome stratification.
View PDF