Document Type : Research Paper

Authors

• Toktam Khatibi ¹
• Niloufar Naddafi ²
• Pooya Mazloomi ³

¹ School of Industrial and Systems Engineering, Tarbiat Modares University, Tehran, Iran

² Faculty of Industrial and Systems Engineering, Tarbiat Modares University, Tehran, Iran

³ Industrial and Systems Engineering, Tarbiat Modares University, Tehran, Iran

Abstract

Glioblastoma multiforme (GBM) represents about 45.6% of primary malignant brain tumors and is marked by rapid growth and resistance to treatment, resulting in a poor prognosis for patients. This study aims to propose a personalized medicine model tailored for patients with GBM with analyzing MRI images and clinical data from 23 patients. Our research encompassed three primary scenarios. In Scenario 1, we constructed a hybrid model combining VIT and Auto-Encoder approaches applied to patient MRI data, achieving an impressive accuracy rate of 96% in determining optimal treatment dosages. For Scenario 2, we introduced Gaussian noise to the MRI images, reflecting real-world conditions, resulting in a drop in model accuracy to 72%. In Scenario 3, we restored the noisy images using advanced techniques, which led to an improved accuracy of 94%. It demonstrates that our proposed scenarios can effectively identify optimal radiotherapy dosages for GBM patients.

Keywords

• glioblastoma multiforme
• radiotherapy
• magnetic resonance imaging
• deep learning
• image noise reduction