Image Segmentation in Multimodal Medical Imaging Using Deep Learning Models

Diagnosis and staging of brain cancer are critical for on-time and appropriate treatment. Generally, manual interpretation of this image data is very time-consuming and subjective. Therefore, the current research aims to find an effective and accurate automated computer-assisted classification of brain tumours using multimodal imaging data, such as CT, MRI, and PET scans. We develop and tune deep learning models, VGG16, VGG19, U-Net, and Gated Recurrent Unit (GRU), to lift the reliability and precision of the analyses of tumor segmentation. The dataset is pre-processed to extract significant features and biomarkers such that spatial and temporal information for each modality is captured. 70% of the data is used for the training of models, and 30% of the data is kept for testing. The highest prediction accuracy was seen in GRU, giving 98.45% in the prediction of brain cancer and its stage, higher than other models. VGG19 followed at a rate of 94.56%. U-Net reached 89.34%, and VGG16 84.5%. Thus, it is beneficial to employ these tests in a clinical, real-time setting, as correct staging and classification frequently improve patient outcomes. Such findings demonstrate that the integration of multi-streamed data into high-order DL models can achieve superior performance in diagnosis and serves as an initial step toward the completely automated investigation of the brain tumor.