Year: 2025
Type: article
Abstract: Extremely dangerous, brain tumors may significantly shorten life expectancy. Because MR scans may provide fine-grained pictures of the afflicted region, the majority of researchers employ them to find malignancies. As of late, through effective data processing, to increase the accuracy of diagnoses, deep learning methods based on AI have emerged. This research examines how well deep transfer learning methods work for precisely identifying brain tumours. Utilising a pipeline for preprocessing enhances the image quality. Morphological methods such as thresholding to trim images, Gaussian blurring to reduce noise, and erosion and dilation for form refinement are all included in this process. Dimensionality reduction is achieved via the use of Principal Component Analysis (PCA), whereas dataset enrichment is achieved through data augmentation. Testing uses 20% of the dataset, while training uses the remaining 80%. GoogleNet and pre-trained ResNet152 extract key elements from the pictures. Following the extraction of these features, the standard machine learning classifiers used for classification include Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), K-Nearest Neighbours (KNN), and Classification and Regression Trees (CART). This research contrasts two pre-trained models for medical image processing. Performance indicators that assess the ultimate categorisation outcomes include accuracy, sensitivity, recall, and F1-Score. ResNet152 beats GoogleNet, according to the findings, with 98.53% accuracy, 96.52% sensitivity, and 97.34% F1 score. Our research emphasises on combining deep learning with conventional machine learning methods for efficient brain processing in order to detect cancers in medical imaging.
Source: Journal of Neonatal Surgery
Institution: Osmania University
Language: English
Subfield: Neurology
Field: Neuroscience
Domain: Life Sciences
Open Access status: diamond
OpenAlex ID: W4409754338