EVENTI
Nell'ambito dei seminari MIA, il 24/2 alle 15,00 in aula Benvenuti, la Dott.ssa Maria Antonietta Pascali, Institute of Information Science and Technologies - CNR, Pisa, terrà il seminario
"Topological Machine Learning: Applications to Raman Spectroscopy".
The advent of machine and deep learning has led to huge advancements in computer vision and data analysis, enabling a shift from handcrafted features to the automatic extraction of meaningful features through representation learning. On the other hand, topological invariants offer informative and computable shape descriptors suitable for differentiating spaces; unfortunately, when applied to real-world data, these descriptors might seem too rigid.
Thanks to the theory of persistent homology (PH), it is possible to use them to conduct intrinsically multiscale analysis. Merging PH and machine learning methods allowed us to design and develop a Topological Machine Learning (TML) pipeline [1], leveraging the informative content of topological features which are different and complementary to those used in deep learning architectures, such as convolutional neural networks. Such a pipeline that associates persistence diagrams to digital data, via the most appropriate ltration for the type of data considered. Using a grid search approach, representation methods and parameters optimal for the classification task assigned are determined. We assessed the performance of our pipeline, and in parallel, we compared the different representation methods, on popular benchmark datasets, showing promising results.
In real-world problems, the pipeline has been exploited in the medical domain for a challenging task: the classification of Raman spectroscopy (RS). RS is based on evaluating the inelastic scattering process in which photons incident on a sample transfer energy to or from molecular vibrational modes. Such information is stored in a spectrum. Instead of focusing the analysis on predetermined peaks or windows in the spectra, current research considers the Raman spectrum as a biochemical signature of the sample. In the biomedical domain (e.g., histopathology and oncology), RS represents a fast and e cient diagnostic tool that has found applications to several kinds of biological samples, including cellular tissues, cell lines and uids, providing practical tools for assessing disease presence and grade. In this context, the TML pipeline achieved convincing results for the grading of chondrosarcoma [2] while in [3,4] it has been used to distinguish the Alzheimer's disease from other neurodegenerative pathologies (Bando Salute 2018 PRAMA project co-funded by the Tuscany Region).
The present contribution will include the description and discussion of strengths and limitations of TML methods applied to the analysis and classification of data from RS in the biomedical domain.
References
[1] F. Conti, D. Moroni, and M. A. Pascali, A topological machine learning pipeline for classification, Mathematics, vol. 10 (2022), no. 17, 3086.
[2] F. Conti, M. D'Acunto, C. Caudai, S. Colantonio, R. Gaeta, D. Moroni, M. A. Pascali, Raman spectroscopy and topological machine learning for cancer grading, Scientific Reports, vol. 13 (2023), no. 1, 7282.
[3] F. Conti, M. Banchelli, V. Bessi, C. Cecchi, F. Chiti, S. Colantonio, C. D'Andrea, M. de Angelis, D. Moroni, B. Nacmias, et al., Alzheimer's disease detection from Raman spectroscopy of the cerebrospinal fluid via topological machine learning, Engineering Proceedings vol. 51 (2023), no. 1, 14.
[4] F. Conti, M. Banchelli, V. Bessi, C. Cecchi, F. Chiti, S. Colantonio, C. D'Andrea, M. de Angelis, D. Moroni, B. Nacmias, M. A. Pascali, S. Sorbi, and P. Matteini, Harnessing topological machine learning in Raman spectroscopy: Perspectives for Alzheimer's disease detection via cerebrospinal fluid analysis, Journal of the Franklin Institute Volume 361, Issue 18, December 2024, 107249.