The complex and time-consuming nature of magnetic resonance imaging (MRI) may make it difficult to autonomously diagnose tumors in the brain, possibly leading to erroneous detection and classification. Identifying brain tumors is a complex process due primarily to relying on multiple modules for a comprehensive evaluation. In response, advances in deep learning have paved the way for automated medical image analysis and diagnostics. Convolutional neural networks (CNNs) are crucial for visual learning and image classification. The current investigation presents a novel approach for data augmentation that is integrated with state-of-the-art models, namely Efficient-NetB0, VGG16, ResNet50, InceptionV3, and MobileNetV2, to accurately classify various types of brain tumors, including glioma, meningioma, and pituitary tumors. The algorithm was subjected to testing utilizing benchmark data from existing literature.
Top1. Introduction
A brain tumor is a pathological state that can be identified by the abnormal proliferation of cells within the cerebral tissues. As per the report by the World Health Organization (WHO), tumors are a substantial cause of death worldwide(Khazaei et al., 2020),(GLOBOCAN, 2020). Brain tumors can be categorized into two distinct groups: benign and malignant. Benign tumors typically exhibit a sluggish growth rate, cannot infiltrate adjacent tissues or organs, and cause a minimal threat to overall health. As the reference (The PDQ Adult Treatment Editorial Board, 2002) states, the permanent abolition of benign tumors is typically achieved through surgical excision. On the contrary, malignant tumors can infiltrate adjacent tissues and organs, potentially leading to severe consequences if not promptly and efficiently managed (The NHS Inform website “Malignant brain tumour,” 2023). The general categories of brain tumors comprise glioma, meningioma, and pituitary. Gliomas are tumors that develop in the glial cells surrounding and sustaining the neurons in the brain, whereas pituitary tumors grow in the pituitary gland (Johns Hopkins Medicine “Gliomas,” 2023)(Mayo Clinic “Pituitary tumors,” 2023). Meningioma is a tumor that develops in the meninges, which are the outermost layers of tissue surrounding the brain and spinal cord between the skull and the brain (Johns Hopkins Medicine “Meningioma,” 2023). Meningioma is generally considered benign (non-cancerous), compared to glioma, which is malignant (cancerous), and pituitary tumors are recognized as benign. The manifestation of symptoms associated with a brain tumor is contingent upon its size and location. These symptoms may encompass headaches, alterations in auditory or visual perception, convulsions, unilateral weakness or numbness, and modifications in personality (MedicineNet “Brain Cancer,” 2022). Furthermore, glioma can cause several symptoms, including but not limited to aphasia, visual disturbances or impairments, cognitive issues, and gait or balance disturbances (Johns Hopkins Medicine “Gliomas,” 2023). Meningioma is known for causing subtle symptoms that gradually increase, such as changes in visual acuity and morning headaches (Johns Hopkins Medicine “Meningioma,” 2023). Pituitary tumors compressing the optic nerve can result in headaches, blurred vision, and diplopia (Mayo Clinic “Pituitary tumors,” 2023).
Hence, it is imperative to differentiate between these categories of tumors to facilitate clinical diagnosis and assessment of treatment efficacy. The proficiency of radiologists plays a pivotal role in the timely detection of brain malignancies. Magnetic resonance imaging (MRI) is a commonly employed modality for identifying and classifying various tumor types. However, it is subject to human interpretation and can pose difficulties in analyzing extensive datasets (Afshar et al. 2019). In diagnosing and treating brain tumors, it is common practice to defer biopsies until after definitive brain surgery, as per existing protocols (Rogers et al., 2015). Developing a comprehensive diagnostic tool for detecting and classifying tumors from magnetic resonance (MR) images is crucial in achieving precise diagnoses and mitigating the need for surgery and subjective assessments.