FusionSegNet: Integrating Modalities for Enhanced Brain Tumor Detection and Segmentation

Improving diagnostic accuracy and striving for thorough analysis is always crucial in medical imaging, Multi Modal Medical Imaging has proved to be a useful approach in this area, providing computational power and accuracy for tasks that humans would otherwise have a tough time doing. This paper investigates the fusion of Magnetic Resonance Imaging (MRI) with Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) or Technetium-99m-Tilmanocept Single-Photon Emission Computed Tomography (TITc-SPECT) scans which are generally used for detecting Gliomas or Metastatic Bronchogenic Carcinoma (Brain Tumors that are formed due to lung cancer). This paper targets a pipeline that will investigate the fusion of Magnetic Resonance Imaging (MRI) with Fluorodeoxyglucose Positron Emission Tomography (FDG-PET) or Technetium-99m-Tilmanocept Single-Photon Emission Computed Tomography (TITc-SPECT) scans which are usually used to detect these brain tumors. The Denoising Diffusion image Fusion Model (DDFM) is used to combine the structural and functional information from these medical scans and the fused results of these images are then used as input to CNN as well as a U-Net model for brain tumor detection and segmentation respectively. The performance of these models is evaluated based on the probability of tumor detection and the accuracy of segmentation. The main objective of this paper is to validate the effectiveness of the DDFM fusion approach for medical scans by demonstrating tumor detection and classification accuracy in the fused multi-modal images. The results of this paper, therefore, hold a potential for improving the clinical decision-making process and making more effective strategies for brain tumor diagnosis and its management through the use of complementary imaging data.