Brain Tumor Detection and Classification Using a Hyperparameter Tuned Convolutional Neural Network

Brain tumor detection using MRI scans when integrated with a deep learning approach can be immensely applied in identifying the tumor at early stages, with minimum medical professional aid. This research paper aims to develop an advanced predictive model that accurately classify brain tumors as benign or malignant using MRI scans. Here, a novel convolutional neural network (CNN) model is proposed to automate tumor detection and improve diagnosis accuracy. The model used a dataset of around 7000 brain cancer data classified into 4 labels which include glioma, meningioma, pituitary, and no tumor. Data wrangling and pre-processing are then applied to unify the images into a single format and remove any inconsistencies. Further the records are segregated into train and test samples with a 70–30 split. The proposed model recorded an optimum accuracy of 94.82%, precision of 94.2%, recall value of 93.7% and f-score metric of 93.9% respectively. In conclusion, the paper concluded that the proposed model can be applied to enhance the precision of both brain tumor diagnosis and prognosis.

I. Introduction

Healthcare based image analytics is a crucial step for early diagnosis of diseases in patients and helps tremendously in finding the right treatment plan for the patient. Life threatening diseases such as brain cancer have a higher mortality rate among females who are 20 years or younger and 40 years old or younger [1]. Currently, brain related cancer ranks as the 10^th most frequent reason for loss of life. Furthermore, individuals diagnosed with brain and CNS cancer have a five-year survival rate of 36%. One of the biggest problems in detecting malignant brain tumors is the diversity in tumor size, shape, location and appearance. So it is challenging to find the accurate measure of the tumor, especially in the beginning stages. Brain cancers such as glioblastoma show high tumor heterogeneity so they are marked by distinct characteristics depending on the genetics of an individual. The human brain is a complex organ with over 100 billion nerves that overlap each other. Detecting a tumor in the brain is complex due to its intricate structure. New advancement in medical technology has made it possible to use machine learning and neural networks to build a prototype that may forecast brain tumors and its type with extreme accuracy. The conventional method of detecting tumors from modalities can be time consuming and erroneous. Automating this step using deep analytics and machine intelligence may enhance the quality of prognosis and diagnosis. By using a publicly available dataset, features retrieved from images, along with data pre-processing techniques and machine learning algorithms are suggested to improve accuracy and decrease diagnostic time. Machine learning can be supervised, where algorithms find the mapping function of input variables and their output labels, or unsupervised, where they rely only on input variables. Two types of supervised learning are classification and regression, while unsupervised learning includes clustering, association techniques and fuzzy c-means.