International Journal of Electrical and Electronics Research (IJEER) Open Access | Rapid and quality publishing Research Article | Volume 10, Issue 4 | Pages 1184-1190 | e-ISSN: 2347-470X 1184 Website: www.ijeer.forexjournal.co.in An Optimized Transfer Learning Based Framework ░ ABSTRACT- Brain Tumor (BT) categorization is an indispensable task for evaluating Tumors and making an appropriate treatment. Magnetic Resonance Imaging (MRI) modality is commonly used for such an errand due to its unparalleled nature of the imaging and the actuality that it doesn't rely upon ionizing radiations. The pertinence of Deep Learning (DL) in the space of imaging has cleared the way for exceptional advancements in identifying and classifying complex medical conditions, similar to a BT. Here in the presented paper, the classification of BT through DL techniques is put forward for the characterizing BTs using open dataset which categorize them into benign and malignant. The proposed framework achieves a striking precision of 96.65%. The proposed framework can be employed to assist physicians and radiologists in validating their initial screening for brain tumor classification. Keywords: Deep Learning, Artificial Intelligence, Image Processing, Transfer Learning. ░ 1. INTRODUCTION Tumor of brain may be portrayed as anomalous as well as uncontrolled growth of cells in the neural structure. Any alarming growth in the brain might affect the human ability; additionally, it may grow into other parts of the body [1]. As per WHO BT signifies below 2% of human cancer [2]. The size of tumor its category along with its position in brain is pivotal in deciding the treatment. Generally, brain surgery is considered as a routine method of handling BT [3]. The most oftentimes happening BTs falls into Glioma categories that consolidate roughly 30% of all Tumors in brain and around 80% of all damaging BTs [4]. In the midst of different clinical developments, MRI produces information about the locale and tumor size. Its function is based on proton activity confined in the magnetic field by varying frequency of radio waves and retrieve their normal state [5] To unequivocally isolate fragile tissues with high exactness MR modality is quite proficient and is progressively receptive to alteration in strength of tissues. The MR modality classify images into T1-weighted (T1-w) which are employed for non-intrusive brain studies as they depict elevated contrast. While, T2-weighted (T2-w) MR images are acceptable for observing the image periphery [6]. The vital pitfall of these images is that BT, Grey Matter (GM) as well as cerebrospinal fluid (CSF) are tied together. Medically, the use of such MR modalities is pivotal in pinpointing tumors nevertheless it may pose some difficulties in sorting out tumorous zones [7]. Consequently, for evaluating the periphery of tumorous tissues counter to a non-tumorous one, use of T1 and T2 weighted contrast modes are important. Tumors of brain are at times mystified for the reason that they continue to be unaltered even with the improvement in their contrast. Successively, the FLAIR images are employed alongside T2-w for showing the non-enhanced BTs [8]. The various MR image types are presented in figure 1. (a) (b) (c) Figure 1: MRI images (a) to (c) - T1, T2, and FLAIR DL is a kind of Artificial Intelligence method that copies the working of a human mind for processing information as well as producing model valuable in settling on appropriate decisions. Deep Learning utilize different layers of non-linear type that are efficient for extrication of image features. The result of each organized layer is the commitment of the accompanying one, and that helps in deliberating data as we jump inside the framework [9]. Convolutional Neural Network (CNN) belongs to the family of DL and the essential interesting point in CNNs are its capability to grasp features and to provide precise exactness instead of customary AI methods by augmenting the training samples and thus prompts to a much robust and exact system [10]. The essential commitment of this proposed work is to present a powerful and robust DL framework utilizing Transfer Learning (TL) strategies for identifying and categorizing BTs by extricating crucial features on a universal image set and then, to explore DL techniques like GoogleNet, ResNet50, and ResNet101 using BT images and apply TL approach on the standard image set to advance a comprehensive performance assessment of features critical for fine tuning of An Optimized Transfer Learning Based Framework for Brain Tumor Classification Manish Kumar Arya 1 and Rajeev Agrawal 2 1 Dr. A.P.J. Abdul Kalam Technical University, Lucknow, India 2 Lloyd Institute of Engineering & Technology, Greater Noida, India, rajkecd@gmail.com *Correspondence: Manish Kumar Arya; manisharya07@gmail.com ARTICLE INFORMATION Author(s): Manish Kumar Arya and Rajeev Agrawal; Received: 24/09/2022; Accepted: 15/12/2022; Published: 20/12/2022; e-ISSN: 2347-470X; Paper Id: IJEER 2409-47; Citation: 10.37391/IJEER.100467 Webpage-link: https://ijeer.forexjournal.co.in/archive/volume-10/ijeer-100467.html Publisher’s Note: FOREX Publication stays neutral with regard to Jurisdictional claims in Published maps and institutional affiliations.