Early Detection of Brain Tumors Using SOLOv3 Algorithm for Enhanced Diagnostic Accuracy

Abstract

The brain tumor is various types occur in the human brain sometime it affects the human quality of life. The deep learning algorithms gives the better detection of the tumor cell with highly positive result in the earlier stage. In the previous work, the customized Segmenting Objects by Locations Vector 3(SOLOv3) algorithm has been approached, this approach gives the better result compared with the previous algorithms. In the medical field the radiographic images are plays the vital role for identifying the disease from the human body at the same time helps to give the proper treatment on time for avoiding the death ratio. There are many automatic image reorganization techniques were developed in deep learning algorithms. Since the proposed idea is to classify the images based on the plasma level and also detect the levels of infection specified as stage1 to stage 5. To use the Magnetic Resonance Images (MRI) for identifying the tissues which present the human organ with the neoplasm type and size. This kind of information’s are helpful to treat the patient on time and also reduce the death rate due this late treatment or detection of the which level the person may affected. The aim of the proposed article is to develop the customized SOLOv3 algorithm with DESNET201 for improved image segmentation and classification. The real time images were taken from the prescribed reputed Neurological diagnosis center in Chennai. Totally 18759 images were collected under all four categories of tumor and non tumor. Which included 13257 images are tumor images under the category of glioma, meningioma and pituitary and remains comes under non tumor category. The implemented model is to customize the final layer of the neural network form with four different classes will give the better result as 91% in the training set and scored 89% as in the phase of test. This improved model that could combine the SOLOv3 and Desnet201 with customized layer classes for extracting the features used to classify the tumor cells with their different types. The tumor may have more than 150 types of tissues, but gradually these four kinds of classes are very dangerous about to increase the death and spreadable to other body organ. These techniques also able to detect the improved automation for the tumor in Indian children and adults.