Efficient Detection and Categorization of Thyroid Nodules from Medical Ultrasound Images

The use of ML and DL in medical imaging for early illness symptom prediction is substantial. Since 2013, DL has been one of the rising trends in general data analysis. With its multiple hidden layers that allow for a high degree of data abstraction, it is an enhancement of artificial neural networks (ANN). One promising approach for computer vision applications is the use of convolutional neural networks (CNNs). Automatic learning of raw data, particularly photographs, is a capability of deep CNNs. Both the capture and interpretation of images are crucial for the correct diagnosis or evaluation of illness. Devices are able to gather data at a rapid pace with improved resolution because to advancements in image acquisition over the last decade. Nevertheless, computer technology has only just started to improve picture interpretation. Since they are subjective and need highly trained doctors, they are often created by radiologist, physicians, and senior doctors. In medical imaging, computerised tools play a crucial role in facilitating results and improving diagnosis. Early detection and categorization of thyroid nodules by visual inspection and manual study of thyroid ultrasonography (USG) images has traditionally been a tedious process. Identifying benign from malignant thyroid nodules requires manual evaluation of thyroid USG images, which may be a time-consuming and laborious process. The medical industry has seen the emergence of many deep learning models, particularly for the categorization of thyroid nodules, thanks to the growth in processing resources and the fast improvement in technology. Clinical interventions and therapies may be more successful when these nodules are detected early. Thus, a growing number of academics are pushing for the use of computer diagnostic systems (CDS) as a means to statistically and scientifically evaluate USG pictures of thyroid nodules. Creating effective models for identifying and classifying thyroid nodules using ML and DL approaches is the main goal of this effort. In the first stage of this project, two hybrid models, ANN-SVM and CNN-SVM, are suggested.