Design and Comparison of Tumor Segmentation Using an ML-Based Clustering Method

A brain tumor is a common syndrome, where a specific set of cells gather and begin to grow inside a human brain, to interfere with the brain s regular function. Though there are a lot of techniques that act as a preliminary test for tumors, an analysis of the MRI scan image is sometimes enough to predict the presence of the tumor. This study aims in finding the best machine learning algorithm that can detect brain tumors with the highest accuracy value. For this purpose, a dataset of images from MRI scans of human brain s is obtained via Kaggle. After that, this dataset is preprocessed using a few techniques. The techniques involve image scaling and color conversion. Two different machine learning models were produced by two different algorithms. The machine learning algorithms used in this work are the U-net and the FPN. The created models are then trained using the preprocessed datasets. The models are then evaluated using a separate set of photos after training. Two metrics, the IoU and dice coefficient, are used to analyze the training and validation results. Although the parameters for both models remain the same during training and validation, it was ultimately determined that the FPN method is more effective at predicting brain cancers. The algorithm’s ultimate IoU value is 0.865, and its dice coefficient is 0.9034. The model is complete because the results were excellent for an image processing model. Real-time photos are used to test the model one more time before it is finished. This analysis’ findings are also deemed to be exceedingly adequate.