Exploring self-supervised deep sparse autoencoders for robust feature selection in radiomics analysis
7 affiliations
- 1Data Science Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 2Radiology Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 3Medical Physics Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 4Genito-Urinary Oncology Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 5Urology Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 6Radiation Oncology Unit, Fondazione IRCCS Istituto Nazionale dei Tumori
- 7Department of Radiation Oncology (Maastro), GROW School of Oncology and Reproduction, Maastricht University Medical Centre+
Abstract
While considerable effort has been devoted to examining how variations in study protocols, acquisition settings, and annotations influence radiomics features, the role of feature selection (FS) methods largely remains unexplored. This study investigates a self-supervised deep sparse autoencoder ensemble (ensembleAE) and a novel Bayesian variant (bayesianAE) for radiomics FS in a small, class-imbalanced framework. Using a cohort of 100 prostate cancer patients under active surveillance, these models were benchmarked against eight classical FS methods. FS stability was assessed both globally and locally in a soft data-perturbation setting. While global stability measured consistency in overall feature ranking, local stability quantified the agreement in selecting top-ranked feature subsets. Among classical methods, wrappers were the least stable, whereas the filter-based Wilcoxon test (WLCX) exhibited high local stability and performance with moderate global stability. Conversely, bayesianAE demonstrated the highest global stability while matching WLCX in local stability and performance. Furthermore, although bayesianAE yielded local stability comparable to that of ensembleAE, it outperformed the latter in terms of global stability with a significantly lower computational burden (95 × faster and consumes 84% less memory). These findings position bayesianAE as a robust alternative to classical FS methods that could support the development of reproducible radiomics signatures.
Related
- Comparative Evaluation of Feature Selection Algorithms for Predicting Radiation-Induced Complications in Head and Neck Cancer Using Multimodal CT and MRI Radiomics
- QuRad: Radiomics Feature Extraction for QuPath
- Dual-domain self-supervised feature alignment via spectral–spatial representation learning for deepfake anomaly detection
- Cross-corpus consensus feature selection for language-invariant Parkinson’s disease detection: a dual-stream acoustic and spectrogram-attention framework
- Learning from pandemics: Feature selection and clustering for robust digital disease nowcasting
BibTeX
@article{seetha2026exploring,
title = {{Exploring self-supervised deep sparse autoencoders for robust feature selection in radiomics analysis}},
author = {Sithin Thulasi Seetha and Antonella Messina and Alessandra Casale and Enrico Garanzini and Chiara Tenconi and Cristina Marenghi and Fabio Badenchini and Barbara Avuzzi and Mario Catanzaro and Nicola Nicolai and Barbara Noris Chiorda and Silvia Stagni and Sergio Villa and Andre Dekker and Tiziana Rancati},
journal = {Scientific Reports},
year = {2026},
month = jul,
doi = {10.1038/s41598-026-62420-7},
url = {https://doi.org/10.1038/s41598-026-62420-7},
note = {CC BY},
}
RIS
TY - JOUR
TI - Exploring self-supervised deep sparse autoencoders for robust feature selection in radiomics analysis
AU - Sithin Thulasi Seetha
AU - Antonella Messina
AU - Alessandra Casale
AU - Enrico Garanzini
AU - Chiara Tenconi
AU - Cristina Marenghi
AU - Fabio Badenchini
AU - Barbara Avuzzi
AU - Mario Catanzaro
AU - Nicola Nicolai
AU - Barbara Noris Chiorda
AU - Silvia Stagni
AU - Sergio Villa
AU - Andre Dekker
AU - Tiziana Rancati
PY - 2026
DA - 2026/07/01
JO - Scientific Reports
DO - 10.1038/s41598-026-62420-7
UR - https://doi.org/10.1038/s41598-026-62420-7
AB - While considerable effort has been devoted to examining how variations in study protocols, acquisition settings, and annotations influence radiomics features, the role of feature selection (FS) methods largely remains unexplored. This study investigates a self-supervised deep sparse autoencoder ensemble (ensembleAE) and a novel Bayesian variant (bayesianAE) for radiomics FS in a small, class-imbalanced framework. Using a cohort of 100 prostate cancer patients under active surveillance, these models were benchmarked against eight classical FS methods. FS stability was assessed both globally and locally in a soft data-perturbation setting. While global stability measured consistency in overall feature ranking, local stability quantified the agreement in selecting top-ranked feature subsets. Among classical methods, wrappers were the least stable, whereas the filter-based Wilcoxon test (WLCX) exhibited high local stability and performance with moderate global stability. Conversely, bayesianAE demonstrated the highest global stability while matching WLCX in local stability and performance. Furthermore, although bayesianAE yielded local stability comparable to that of ensembleAE, it outperformed the latter in terms of global stability with a significantly lower computational burden (95 × faster and consumes 84% less memory). These findings position bayesianAE as a robust alternative to classical FS methods that could support the development of reproducible radiomics signatures.
SN - 2045-2322
ER -