Enhancing Brain Tumor Diagnosis: Utilizing ResNet-101 on MRI Images for Detection

Abstract

Brain cancer is an increasingly grave and oftentimes severely painful condition that has attracted a lot of attention. Although brain cancer is rare and the disease is less prevalent than many other cancer kinds, 60% of cases survive within one year of diagnosis, compared to 30% of cases- that barely survive five years. This is a moderately low survival rate. The survival rate has somewhat increased over the past decade for patients discovered in the earlier stages. Nonetheless, it appears that the overall number of persons with brain cancer will continue to climb shortly due to the aging population. Identifying symptomatic individuals at the initial stage is a crucial public health approach to achieving this goal. One such model is our proposed model, which utilizes ResNet-101 architecture in the core network. The proposed model is trained using magnetic resonance images (MRIs) of the brain and utilizes transfer learning to improve model performance. The ResNet-101 architecture enables the use of residual blocks and skips connections to address the vanishing gradient problem and improve model accuracy. A collection of 3060 brain MRI data is used to evaluate the suggested system and achieves an accuracy of 97% in classifying tumors. This methodology may increase the precision and effectiveness of brain cancer identification, aiding in early diagnosis and treatment.