COMPUTER-BASED NODULE MALIGNANCY RISK ASSESSMENT IN THYROID ULTRASOUND IMAGES

摘要

This paper presents a computer-based approach for detection, de-lineation, and malignancy risk assessment of thyroid nodules in ultrasound (US) images. The proposed approach is automatic and integrates processes for: the thyroid gland boundaries detection, the detection of nodular lesions within the thyroid gland, the delineation of the detected nodules, and the classification of thyroid nodules according to malignancy risk. These processes embed textural and shape feature vectors derived from the US images, as well as state-of-the-art medical image analysis and pattern recognition tools. The obtained classification performance, which is associated with automatic malignancy risk assessment, was evaluated by means of the receiver operating characteristic (ROC), demonstrating an area under curve (AUC) equal to 0.93. The quantification of the results shows that the proposed approach: (1) contributes to the objectifi-cation of the diagnostic process by the utilization of explicit image features, whereas it can provide the diagnosticians with a second opinion, (2) is applicable in clinical practice and could contribute to the reduction of false medical decisions.