Novel hybrid approach based on combination of textural features and clinical parameters for reliable prediction of thyroid malignancy

Abstract

The dramatic increase in thyroid cancer, particularly among the younger population demands development of an automated decision support system for timely and reliable prognosis of the disease so as to facilitate improved chances of recovery in the subjects. While numerous methods are already reported by the researchers for the detection of Thyroid malignancy, the most crucial parameter of Thyroid Malignancy Index (TMI) has received very less attention. TMI is of paramount importance in diagnosis and treatment of the patients having malignant thyroid and its consideration is therefore inevitable. This research aims to develop an automated and a reliable decision support system for detection of thyroid malignancy. The proposed hybrid approach incorporates a novel combination of texture features and clinically observable parameters to initially identify a malignant thyroid tumor using support vector machines and further predicts its TMI value, thus exhibiting a performance like a trained radiologist. Publically available database comprising of 99 cases and 134 ultrasound images are used to validate the superiority of the proposed approach. Apt consideration and reliable prediction of the TMI in this research makes the designed approach novel and marks a mega leap towards its practical deployment in the clinical environment.