Brain Tumor MRI Segmentation Using Deep Instance Segmentation with Bioinspired Optimization Algorithm

Brain tumor (BT) segmentation plays a major part in medical image processing. Earlier detection of BTs is an important task in improving treatment possibilities and increasing the patient survival rate. Manual segmentation of BTs for cancer detection, from the abundance of MRI images generated in medical routine, is a time-consuming and difficult task. It is necessary for the automated segmentation of BT images. BT segmentation methods dependent on classical machine learning (ML) and image processing are not ideal sufficient among the existing brain segmentation method. Thus, the DL-based brain segmentation method is more commonly applied. In the BT segmentation technique based on DL, the convolution network technique has a better brain segmentation effect. The deep convolutional network method has the problem of great loss of information and a huge number of parameters in the encoder and decoder processes. With this motivation, this article presents a new Brain Tumor Segmentation using Al‐Biruni Earth Radius Optimizer with Deep Learning (BTS-AERODL) technique on MRI images. The major intention of the BTS-AERODL technique is to segment the affected BT regions in the MRI images. In the presented BTS-AERODL technique, an initial stage of pre-processing takes place in two levels: Non-Local Means (NLM) filter-based noise removal and CLAHE-based contrast enhancement. In addition, the BTS-AERODL technique applies the Inception-CBAM Unet++ (ICUnet++) method for the image segmentation process. Finally, the hyperparameter tuning of the ICUnet++ model takes place by the use of the AERO algorithm. The experimental validation of the BTS-AERODL model is validated on a benchmark brain MRI datasets. The wide results stated the enhanced results of the BTS-AERODL model over other approaches.