Mario Pascual González

Biomedical Imaging Researcher

University of Málaga

Trying to understand how silicon-based machines can understand the world. Applications to Biomedical Imaging, Graph Representation, and more.

Selected Publications

Preprint

SLIM-Diff: Shared Latent Image-Mask Diffusion with Lp loss for Data-Scarce Epilepsy FLAIR MRI

M. Pascual-González, A. Jiménez-Partinen, R. M. Luque-Baena, F. Nagib-Raya, E. López-Rubio

arXiv preprint arXiv:2602.03372, 2026

Paper Code Zenodo PyPI
Personal Comments

Preprint, sent to the IEEE ICIP 2026 conference. The code provides a PyPI package that allows you to easily train your own diffusion model with the SLIM-Diff architecture, which is especially designed for data-scarce 2D MRI synthesis tasks. The paper also provides empirical evidence that the manifold hypothesis holds for pixel-space generation of medical images, a claim that 'One-step Latent-free Image Generation with Pixel Mean Flows' (Yiyang Lu et.al) made, with theoretical grounding.

Q1

Hyperparameter optimization of YOLO models for invasive coronary angiography lesion detection and assessment

M. Pascual-González, A. Jiménez-Partinen, E. J. Palomo, E. López-Rubio, A. Ortega-Gómez

Computers in Biology and Medicine, vol. 196, pp. 110697, 2025

Paper Code
Personal Comments

My first scientific publication. Leaving the statistical analysis aside, the code provides an easy tool that is able to integrate the hyperparameter optimization of any YOLO variant using the Optuna framework. You only have to define your trainer at optimization/engine/trainers. The script optimization/engine/hpo.py will catch your trainer class and optimize it with the configuration you defined (see example at optimization/cfg/files/picasso/yaml), where you must write your trainer class in the model_source entry.

View all publications →

Recent Posts

View all posts →