A Deep Learning Approach For Precise Classification Of Nasal Polyps: Distinguishing Antrochoanal Polyps And Ethmoidal Polyps

Abstract

Nasal polyps are benign growths that occur within the sinuses or nasal passages. Because of their interconnected characteristics and diverse appearance, they present significant challenges in terms of diagnosis and treatment. For these polyps to be effectively managed clinically, they must be accurately classified. However, manual examination-based traditional approaches are complicated and prone to error. In order to create an automated system that accurately classifies nasal polyps and can differentiate between antrochoanal polyps (AP) and ethmoidal polyps (EP), this work makes use of progressive deep learning techniques. Several advanced convolutional neural networks (CNNs) are used in the study, including ResNet-50, VGG16, PolyScanCNN (a customised CNN model), and EfficientNetB0. A large dataset of nasal MRI images, which underwent intensive preprocessing to improve picture quality and augment the data, served as the training set for these models. The models were assessed and compared using performance indicators such F1-score, recall, accuracy, and precision. The ResNet-50 model achieved a 96% accuracy rate, demonstrating its effectiveness in identifying nasal polyps. The study shows how well these models work in identifying nasal polyps, with ResNet-50 outperforming the other models in the majority of evaluation measures. Gradio was used to create an intuitive web-based interface that allows real-time nasal polyp picture categorization and visualization, hence promoting clinical adoption. This research highlights how deep learning can improve otolaryngology diagnostic efficiency and accuracy. It seeks to assist physicians in making wise decisions by offering a trustworthy tool for the classification of nasal polyps, ultimately leading to better patient outcomes. The results indicate that the field of medical image analysis can be greatly advanced by implementing advanced AI-driven solutions.