الفهرس 
AcknowledgmentOnly 14 pages are availabe for public view
 |  | from | 84 |  |  |
| | | from | 84 | | |
Abstract
Thyroid cancer tends to have a relatively high survival rate and is often treatable if detected early. However, its prevalence and impact can vary by country due to factors like healthcare access, screening practices, and environmental factors. The ultrasound images can detect abnormal changes in the structure or size of the thyroid gland. This includes the presence of nodules or growths within the thyroid tissue. The present study presents a framework to classify thyroid cancer cells using ultrasound images into two main specified cell types using deep learning algorithms. The digital database of Thyroid Images (DDTI) is used in this investigation. Depending on their internal composition, echogenicity, margins, calcifications, and TI-RADs, experts categorize the cells into two main distinct types which are benign and Malignant cells. Introducing a pipeline to improve diagnostic accuracy by experimenting with Keras application models and choosing the appropriate model result is the main contribution. The study concluded that deep learning could improve ultrasound screening classification results. The most suitable algorithms for this application appear to be VGG16, MobileNet, and InceptionResNetV2. The VGG16, MobileNet, and InceptionResNetV2 have the highest accuracy of 99.9%. These results support the proposed framework as a reliable diagnostic tool for thyroid cancer.