Automated brain tumor malignancy detection via 3D MRI using adaptive-3-D U-Net and heuristic-based deep neural network

Using the 3D image from public benchmark sources, the experiment is initiated with pre-processing using skull stripping and contrast enhancement. Further, the segmentation of tumor region is performed by the Adaptive-3-D U-Net (A-3D-U-Net) utilized for the hybridized Butterfly Optimization Algorithm (BOA), and Tunicate Swarm Algorithm (TSA) termed to as Butterfly–Tunicate Swarm Algorithm (B-TSA). The optimal segmentation of tumors is based on solving the multi-objective solution concerning “Structured Similarity Index (SSIM), Mean Square Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Dice Coefficient”. From the segmented tumor region, the numerical features such as “mean, standard deviation, entropy, skewness, kurtosis, energy, contrast, inverse difference moment, directional moment, correlation, coarseness, and texture features like Local Ternary Pattern (LTP), and Local Tetra Pattern (LTrP)” are extracted. In the final stage, the detection of malignancy is performed by heuristic-based deep neural network (HDNN) using the same proposed B-TSA for the parameter optimization. The findings of applying the suggested methodology to 3D-MRI images from the Decathlon dataset demonstrate that the suggested technique is comparable to conventional methods for brain tumor segmentation.