Medical Image Analysis for Disease Detection and Categorisation Using Deep Learning Algorithm

Abstract

Medical image classification is crucial for clinical treatment and educational purposes. The conventional approach has
attained its performance limit. Utilizing them requires considerable time and effort to extract and select categorization
characteristics. The deep neural network (DNN) is an innovative Deep Learning (DL) technique that has demonstrated
its efficacy in many categorization problems. The Convolutional Neural Network (CNN) excels, achieving superior
outcomes in diverse picture categorization tasks. Compiling medical picture collections is challenging due to the
requisite professional competence for accurate labeling. This paper investigates the application of a CNN algorithm
on a chest X-ray database for pneumonia classification. Three methodologies are assessed by experimentation. This
text describes a linear Support Vector Machine (SVM) classification utilizing local rotation and orientation-invariant
features, employing transfer learning on two convolutional neural network examples: Visual Geometry Group
(VGG16) and InceptionV3, as well as a capsule network trained from the ground up. Enhancing data is a preprocessing
technique utilized across all three approaches. The experimental results indicate that data enrichment is usually
successful in improving the efficiency of all three methods. Transfer learning is a more effective classification
technique for short datasets than an SVM utilizing Oriented Fast and Rotated Binary (ORB) resilient dependent
elementary variables and capsule networks. In machine learning, it is crucial to retrain particular features on a new
target database to enhance efficiency. The second critical factor is an appropriate network connection corresponding
to the dataset's size.