Thyroid Cancer Diagnosis from Sonogram Images using Machine Learning

Introduction :

Thyroid is a gland shaped like a small butterfly, which is found inside the lower front of the neck. It releases hormones that direct many functions in the human body and also controls our metabolism. Nowadays Thyroid diseases are very common and treated as a lifestyle disease. Thyroid diseases are mainly classified into two types:Hyperthyroidism and Hypothyroidism. These two conditions often lead to many other diseases like diabetics, heart and kidney problems etc. Yet another thyroid problem is thyroid cancer, which sometimes show no symptoms. Thyroid Cancer develops when cells change or mutate. The abnormal cells begin multiplying in the thyroid gland and, once there are enough of them, they form a tumour. If it’s caught early, thyroid cancer is one of the most treatable forms of cancer. All thyroid nodules do not lead to thyroid cancer.The four types of thyroid cancers are Papillary thyroid cancer, Follicular thyroid cancer, Medullary cancer and Anaplastic thyroid cancer. For thyroid cancer diagnosis, ultrasound scanning of Thyroid nodules along with blood hormone studies or CT scan results and blood chemistry studies are the common methods. Ultrasound scanning plays a vital role in diagnosis of Thyroid Cancer. But in ultrasound images, the cancer regions are usually blurred, with vague margin and irregular shape. So identification of cancers from the images need expertise and accuracy.

Objective of the Research :

All these studies reveal that a lot of research is being carried out in detection and prediction of thyroid cancer using different modalities. Accurate determination of thyroid nodules and classification of whether malignant or benign could reduce the risk factor to a great extent. The proposed work aims to focus on developing a novel methodology for Cancer diagnosis of Thyroid nodules from Sonogram images using deep learning methods with improved precision and accuracy. It is also proposed to build a prototype to combine detection and classification using deep learning methods. In this study we would like to effectively classify the malignant thyroid nodules and identify the specific type of thyroid through the feature extraction of the sonogram images of thyroid nodules. Sonographic characteristics of thyroid nodules can be used for the classification and diagnosis of thyroid cancers.