MaLGa logoMaLGa black extendedMaLGa white extendedUniGe ¦ MaLGaUniGe ¦ MaLGaUniversita di Genova | MaLGaUniversita di Genova

Computer vision for biomedical data challenges

dgt book

We explore machine learning and computer vision solutions to challenges typical of biological and medical data, including learning with limited or no labels and interpretability. We design and implement methods for the solution of a broad range of biomedical data related applications.


  • Francesca Odone - DIBRIS, Università di Genova

  • Nicoletta Noceti - DIBRIS, Università di Genova

  • Vito Paolo Pastore - DIBRIS, Università di Genova

Methods: learning without labels for biological fine-grained data

Supervised learning has been massively applied to the solution of different computer vision tasks related to biological data, spanning from classification to segmentation. However, annotations do not come for free, and a lot of resources is generally needed to provide accurate and unbiased labels. Moreover, some biological applications may have not uniquely defined labels.

In this project, we investigate representation learning and unsupervised learning methods to provide efficient and accurate mappings for fine-grained biomedical dataset.

Collaboration with


P. D. Alfano, M. Rando, M. Letizia, F. Odone, L. Rosasco and V. P. Pastore, Efficient Unsupervised Learning for Plankton Images, 2022 26th International Conference on Pattern Recognition (ICPR), 2022, pp. 1314-1321.

Applications: Environmental monitoring with plankton

The project aims in using computer vision and machine learning to exploit plankton microorganisms as biosensors, capable of detecting perturbation of an aquatic environment. Reaching this objective requires establishing a morphological and behavioral baseline for healthy plankton, detecting environmental perturbations as deviations from such a baseline. A fundamental challenge is that plankton microorganisms have an enormous intraand inter-species genetic and phenotypic diversity, that coupled with the limited amount of large annotated datasets, makes it hard to obtain a complete representation of this important class of organisms. In this context, we study and develop novel and efficient unsupervised learning approaches to be applied to the task at hand while evaluating different transfer-learning scenarios to exploit existing datasets for knowledge transfer.

Collaboration with


Vito Paolo Pastore, Nimrod Megiddo, and Simone Bianco. An Anomaly Detection Approach for Plankton Species Discovery. In Image Analysis and Processing – ICIAP 2022: 21st International Conference, Lecce, Italy, May 23–27, 2022, Proceedings, Part II. Springer-Verlag, Berlin, Heidelberg, 599–609.

Pastore, V.P., Zimmerman, T.G., Biswas, S.K. et al. Annotation-free learning of plankton for classification and anomaly detection. Sci Rep 10, 12142 (2020).

Vito P. Pastore, Thomas Zimmerman, Sujoy K. Biswas, Simone Bianco, Establishing the baseline for using plankton as biosensor, Proc. SPIE 10881, Imaging, Manipulation, and Analysis of Biomolecules, Cells, and Tissues XVII, 108810H (4 March 2019).

Methods: Explainable machine learning for biomedical data analysis

Deep learning has been successfully and extensively applied to biomedical data at different scales, from single cells to entire organs or systems. Despite the high accuracy obtained for different tasks, from segmentation to classification and diagnosis, neural networks intrinsically lack interpretability, potentially causing a bottleneck in the clinical application. The project aims in designing and exploring the application of explainability methods and tools to provide interpretable features for biomedical data analysis.