Título: Exploring Advanced Deep Learning Paradigms for Precise Brain Tumor Categorization

Autor(es): IMBAQUINGO ESPARZA DAISY ELIZABETH, JACOME LEON JOSE GUILLERMO, MARCELO ZAMBRANO‑VIZUETE, MIGUEL BOTTO‑TOBAR

Fecha de publicación: 15-oct-2024

Resumen: Current developments in medical image processing have relied on deep learning. One potential use for deep learning is to improve brain tumor categorization. The primary goal of this work is to create deep learning models that can detect brain cancer in MRI data. Convolutional neural networks (CNNs) and transfer learning are two recently discovered alternatives to conventional tumor classification methods. We designed these approaches to address the concerns mentioned above. One of the product's major flaws is its inability to recognize and make broad judgments about human qualities. The development of deep learning models for brain tumor detection facilitated achieving the stated goal. In addition, we analyzed CNN designs and transfer learning approaches, investigated data augmentation strategies to increase model performance, and examined classic machine learning processes. These similarities happened together. Deep learning models, particularly CNNs, outperform more traditional approaches in terms of accuracy, durability, and processing resource efficiency. In this scenario, it is clear how deep learning may help with the identification and treatment of brain cancers. This research project aims to provide a unique deep learning technique for brain tumor categorization. This approach combines many feature extraction methods with modern model designs. Overall, the suggested method outperformed the alternatives. This collection provides several unique style alternatives. ResNet, VGG, DenseNet, and many more architectures are among the many that fall into this category. Trial results comparing several deep learning approaches corroborated these conclusions. When compared to previous models that followed the suggested technique, the new model performed better on all examined features. The following characteristics were considered: AUC-ROC, recall, accuracy, precision, F1 score, and F1. The findings revealed an F1 score of 0.90, accuracy, precision, and area under the receiver operating characteristic curve (AUC-ROC) of 0.95, a 0.88 recall rate, and 0.90 accuracy and precision. According to the study results, the proposed strategy enhances the reliability of tumor classification. According to the research, complicated feature extraction and model update procedures are required for improved classification performance. These discoveries have inspired changes in clinical practice, perhaps leading to improved patient outcomes and more accurate diagnoses. The data may yield insights beyond these two assumptions. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 202

Palabras clave: Brain tumor classification, Convolutional neural networks, Deep learning · Medical image analysis, Neuroimaging

DOI: https://doi.org/10.1007/s42979-024-03228-y

ISSN: 2662995X

Tipo publicación: Artículo