Enhanced Tumor Classification via Radiomics and Deep Feature Generation using Neural Network: Integrating Radiomics with CNN-Based Deep Features

Document Type

Conference Proceeding

Source of Publication

2025 Medical Technologies Congress (TIPTEKNO)

Publication Date

10-26-2025

Abstract

Glioblastoma (GBM) is the most aggressive and prevalent malignant brain tumor in adults. Accurate preoperative differentiation between GBM and other gliomas remains a critical challenge due to imaging overlaps and tumor heterogeneity. Radiomics has emerged as a powerful tool to extract quantitative imaging features; however, most existing approaches rely on traditional radiomic features derived from limited modalities or tumor regions. In this study, we have proposed a novel hybrid pipeline that integrates traditional radiomic features with deep features extracted using a convolutional neural network (CNN). We used multimodal magnetic resonance images and four pre-segmented tumor subregions (necrosis, edema, enhancing tumor, and whole tumor) from 185 brain tumor patients publicly available in The Cancer Imaging Archive (141 GBM, 44 non-GBM). A total of 1544 radiomic features were extracted using the PyRadiomics library for each patient. After applying FeatureWiz for feature selection, a Support Vector Machine classifier was trained using only selected traditional radiomic features and both selected traditional radiomic features and CNN-based deep features. Our results demonstrate that combining traditional radiomic and deep features significantly improves classification performance. Validation accuracy increased from 0.838 to 0.948, F1-score from 0.558 to 0.948, and recall from 0.436 to 0.950. These improvements indicate better generalization and robustness of the hybrid model. This study highlights the potential of integrating deep and traditional radiomic features for glioblastoma classification. Our hybrid pipeline addresses limitations of existing radiomics-only approaches and provides a promising direction for non-invasive tumor characterization in neuro-oncology.

Publisher

IEEE

First Page

1

Last Page

4

Disciplines

Computer Sciences | Medicine and Health Sciences

Keywords

Radiomics and Machine Learning in Medical Imaging, Brain Tumor Detection and Classification, Glioma Diagnosis and Treatment

Indexed in Scopus

no

Open Access

no

Link to Full Text

Share

COinS