Abstract
Hyperspectral imaging (HSI) provides rich spectral information for biomedical image analysis but exhibits high dimensionality, which increases computational complexity and introduces redundant information. This paper investigates the impact of dimensionality reduction on hyperspectral overlapping chromosome segmentation . Three linear methods, namely Principal Component Analysis (PCA), Non-negative Matrix Factorization (NMF), and Factor Analysis (FA), and three nonlinear methods, namely Kernel Principal Component Analysis (Kernel PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and Isomap, are evaluated. Each method projects the original hyperspectral cube into a three-component representation used as input to 2D U-Net architecture. The resulting models are compared with a 2D U-Net using RGB images, as well as 2D and 3D U-Nets operating on the complete hyperspectral cube. Experiments are conducted using five-fold cross-validation and evaluated using the Dice coefficient and Intersection over Union (IoU). Paired Wilcoxon signed-rank tests are performed to assess statistical significance. The results show that dimensionality reduction significantly improves segmentation performance compared with both RGB images and the complete HSI cube. Although Isomap achieved the highest average Dice and IoU scores, pairwise statistical analyses revealed no significant differences between most dimensionality reduction methods. These findings demonstrate that dimensionality reduction is an effective preprocessing strategy for hyperspectral overlapping chromosome segmentation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMERGE_028.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=ksGB8VLENR
BibTex
@InProceedings{ZefNaj_Linear_MICCAISAT2026,
author = { Zeffate, Najoua AND Bechar, Mohammed El Amine AND Guyader, Jean-Marie AND Douet-Guilbert, Nathalie AND El Bouz, Marwa AND Troadec, Marie-Bérengère},
title = { { Linear and Non-Linear Dimensionality Reduction for Hyperspectral Overlapping Chromosome Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17266},
month = {pending},
page = {pending}
}
