![[2602.18119] RamanSeg: Interpretability-driven Deep Learning on Raman Spectra for Cancer Diagnosis](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18119] RamanSeg: Interpretability-driven Deep Learning on Raman Spectra for Cancer Diagnosis
Summary
The paper presents RamanSeg, an interpretable deep learning model for analyzing Raman spectra in cancer diagnosis, achieving significant accuracy improvements over traditional methods.
Why It Matters
This research addresses the limitations of histopathology by introducing Raman spectroscopy as a faster, stain-free alternative for cancer diagnosis. The development of an interpretable model enhances trust and usability in clinical settings, potentially transforming diagnostic practices.
Key Takeaways
- RamanSeg achieves a mean foreground Dice score of 80.9%, outperforming previous models.
- The model offers a trade-off between interpretability and performance through two variants.
- Raman spectroscopy provides a stain-free method for cancer diagnosis, streamlining the process.
Electrical Engineering and Systems Science > Image and Video Processing arXiv:2602.18119 (eess) [Submitted on 20 Feb 2026] Title:RamanSeg: Interpretability-driven Deep Learning on Raman Spectra for Cancer Diagnosis Authors:Chris Tomy, Mo Vali, David Pertzborn, Tammam Alamatouri, Anna Mühlig, Orlando Guntinas-Lichius, Anna Xylander, Eric Michele Fantuzzi, Matteo Negro, Francesco Crisafi, Pietro Lio, Tiago Azevedo View a PDF of the paper titled RamanSeg: Interpretability-driven Deep Learning on Raman Spectra for Cancer Diagnosis, by Chris Tomy and 11 other authors View PDF HTML (experimental) Abstract:Histopathology, the current gold standard for cancer diagnosis, involves the manual examination of tissue samples after chemical staining, a time-consuming process requiring expert analysis. Raman spectroscopy is an alternative, stain-free method of extracting information from samples. Using nnU-Net, we trained a segmentation model on a novel dataset of spatial Raman spectra aligned with tumour annotations, achieving a mean foreground Dice score of 80.9%, surpassing previous work. Furthermore, we propose a novel, interpretable, prototype-based architecture called RamanSeg. RamanSeg classifies pixels based on discovered regions of the training set, generating a segmentation mask. Two variants of RamanSeg allow a trade-off between interpretability and performance: one with prototype projection and another projection-free version. The projection-free RamanSeg outperformed a U-Net ...