A Novel Multi-Task Deep Learning Framework for Glioma Segmentation, Survival Prediction, and Recurrence Detection
International Journal of Drug Delivery Technology · 03.07.2026 · 16 (61s)
DOI: 10.25258/ijddt.16.61s.36 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Drug Delivery Technology, online erschienen am 03.07.2026. Verfasst von Kommalapati Naveen Kumar und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Brain tumour segmentation and persistence with a survival prediction are the most key important of medical issues and problems. To need computer-based AI models to transaction to deal with these brain tumour issues and problems. When it comes to brain tumour segmentation with people often use learning models like CNN and UNet. These are methods but they have some limitations. They do not work well with brain tumours, and they have trouble, with spatial coherence and time variance. Brain tumours segmentation is an issue that requires better solutions. To address these challenges, in this work, we integrate a Modified SOFM, V-Net, Recurrent U-Net, and Cox Proportional Hazards (CoxPH) survival analysis to build a comprehensive tumor analysis framework. V-Net: A 3D convolutional neural network adopts volumetric features from multimodal MRI (T1, T1ce, T2, and FLAIR) to enhance the segmentation performance. Recurrent U-Net is more appropriate for the tracking of the development or relapse of the tumors, since it improves the segmentation of using recurrent connections to capture temporal dependencies. The segmented tumor features returned by V-Net are passed to the CoxPH model, which predicts survival probabilities based on crucial parameters such as tumor volume, shape and some demographics of patients (such as age). CoxPH provides a statistically valid means to predict survival time by integrating clinical and imaging information to prognosticate at a more personalized level. Predicting the occurrence of recurrence is important for long-term follow-up of the patient, and survival analysis alone is inadequate. Recurrent U-Net with spatialtemporal learning of sequential imaging data enhances A.M. Noël/Magn. Reson. Imaging 64 (2020) 41–48 47 precision of FM segmentation and offers potential for earlier detection of tumour recurrence. This capability is further enhanced by a modified SOFM that clusters extract tumor features for finding recurrence patterns and assures a more accurate follow-up strategy.
Zitieren
APA
Kumar, K. N., Dhanalakshmi, P., & Baseer, K. (2026). A Novel Multi-Task Deep Learning Framework for Glioma Segmentation, Survival Prediction, and Recurrence Detection. International Journal of Drug Delivery Technology, 16(61s). https://doi.org/10.25258/ijddt.16.61s.36
Vancouver
Kumar KN, Dhanalakshmi P, Baseer K. A Novel Multi-Task Deep Learning Framework for Glioma Segmentation, Survival Prediction, and Recurrence Detection. International Journal of Drug Delivery Technology. 2026;16(61s). doi:10.25258/ijddt.16.61s.36
BibTeX
@article{Kumar2026,
author = {Kumar, Kommalapati Naveen and Dhanalakshmi, P. and Baseer, K.K.},
title = {A Novel Multi-Task Deep Learning Framework for Glioma Segmentation, Survival Prediction, and Recurrence Detection},
journal = {International Journal of Drug Delivery Technology},
year = {2026},
volume = {16},
number = {61s},
doi = {10.25258/ijddt.16.61s.36}
}
EUVIMED