International Journal of Electrical and Computer Engineering (IJECE) Vol. 15, No. 3, June 2025, pp. 3226~3237 ISSN: 2088-8708, DOI: 10.11591/ijece.v15i3.pp3226-3237  3226 Journal homepage: http://ijece.iaescore.com Robust deep learning approach for accurate detection of brain tumor and analysis Lanke Pallavi, Thati Ramya, Singupurapu Sai Charan, Sirigadha Amith, Thodupunuri Akshay Kumar Department of Computer Science and Engineering, B V Raju Institute of Technology, Narsapur, India Article Info ABSTRACT Article history: Received Jul 15, 2024 Revised Dec 10, 2024 Accepted Dec 19, 2024 Usually, one of the foremost predominant and intricate therapeutic conditions. As broadly perceived, brain tumors are among the foremost significantly harmful circumstances that can radically abbreviate a person’s life expectancy. Various methods are lacking for observing the assortment of tumor sizes, shapes, and areas. When merged with strategies of profound learning, generative adversarial networks (GANs) are competent of catching the measurements, areas, and structures of tumors. Profound learning frameworks will move forward upon the shortage of datasets. It can moreover progress photographs with determination. Classifying and partitioning brain tumors productively is significant. GANs are used in conjunction with an overarching learning handle. A profound learning design called NeuroNet19, could be an intercross of visual geometry group (VGG19) and inverted pyramid pooling module (IPPM) which is utilized to recognize brain tumors. It is clear that, NeuroNet19 employments the foremost exact technique in comparison to all models (DenseNet121, MobileNet, ResNet50, VGG16). The exactness examination gave a Cohen Kappa coefficient of 99% and a F1-score of 99.2% Keywords: DenseNet121 Generative adversarial networks Inverted pyramid pooling module Medical imaging brain tumor meningioma pituitary tumor Magnetic resonance imaging NeuroNet19 glioma This is an open access article under the CC BY-SA license. Corresponding Author: Thati Ramya Department of Computer Science and Engineering, B V Raju Institute of Technology Narsapur, Telangana, India Email: ramyathati1102@gmail.com 1. INTRODUCTION Brain tumors may be defined as a growing abnormality of brain cells. Such a biological complex architecture requires quite a challenging medicine and also the possibility of impairing brain functions [1]. They may exert pressure on surrounding tissues, hence eventually leading to intracranial pressure and fluid accumulation [2]. Since they could come in any part of the brain and may vary in size and severity, diagnosis and treatment turn out quite challenging. Brain tumors are basically classified into two categories according to their origin: benign and malignant [3]. These benign types are generally non-aggressive, characterized by slow growth and pose relatively less threat to health [4]. On the other hand, malignant tumors are aggressive; they can invade surrounding areas as well as spread to almost all parts of the body and pose a serious threat to human health, if not treated with excellent effectiveness [5]. Thus, early brain tumor detection is important for patients improvement through the provision of early medical therapy. Many techniques have been designed for the detection of brain tumors, all of which have an application and a drawback [6], [7]. However, in most conventional image diagnostics, the models rely on machine learning- based models that are easily confused with the inherent complexity of brain tumors since of variations in growth patterns and characteristics [8], [9]. Recent research has focused on deep learning techniques [10] that