Radiomics-GenAI: a novel model for laryngeal lesion classification
Year: 2026
Type: article
Abstract: Narrow-band imaging (NBI) enhances mucosal and vascular contrast and has become an important modality for the early detection of laryngeal cancer. Although radiomics enables the extraction of high-dimensional quantitative features from NBI images for computer-assisted diagnosis, the limited availability and heterogeneity of annotated datasets constrain the development and generalizability of machine-learning models. We propose a novel generative AI framework that improves radiomics-based classification. Instead of generating synthetic medical images, the proposed framework trains generative models directly in the radiomic feature space to learn realistic feature distributions and generate valid synthetic radiomic representations. The framework was evaluated with multiple machine-learning classifiers on a laryngeal cancer diagnosis task using a dataset comprising 11,144 NBI images from 210 adult patients, including 180 patients in the development set and 30 patients in an independent internal held-out test set. Performance was evaluated on an independent unseen test set using , accuracy, precision recall and F1-score. Generative augmentation consistently improved classification performance across all evaluated classifiers. Among the evaluated generative models, Generative Adversarial Networks achieved the most consistent improvements. Radiomics-GAN yielded the highest Median value of F1-scores of 0.851, outperforming the SMOTE-balanced baseline model without GenAI (F1-score: 0.806) on the test set. By enriching the radiomic feature manifold without generating synthetic medical images, the proposed framework effectively mitigates sample scarcity, improves classifier performance, and supports the development of more robust AI-assisted tools for laryngeal cancer diagnosis. Future work will focus on external validation and conditional generative models tailored to lesion subtypes and imaging acquisition variability.
Source: BMC Medical Imaging
Authors: Haiyang Wang, Zhihui Xie, Luca Mainardi
Institution: Politecnico di Milano
Language: English
Subfield: Radiology, Nuclear Medicine and Imaging
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,090
OpenAlex ID: W7223464136
FWCI:
Cites:
Cited by:
Related to: