Classification of the Thyroiditis based on characteristic sonographic textural features and correlated histopathology results

Abstract

Thyroiditis is a health disorder and it refers to “inflammation of the thyroid glands”. Once a thyroid nodule has been detected (or suspected), the first test that is routinely being performed is the fine needle aspiration (FNA) biopsy (invasive). This test result is helpful in the classification of the nodule being benign or malignant with the aid of bio-markers. Another common test is the Ultrasound imaging (non-invasive). But, due to the inherent spatial limitations within the Ultrasound image, distinguishing the different pathological conditions related to Thyroiditis are challenging. The aim of this study is to classify multiple pathological conditions related to Thyroiditis by analyzing various textural image features extracted from ultrasound images. The thyroid Ultrasound images are retrospectively collected (with biopsy results) from a private scan center in Chennai, India. The image database contains thirty five Adenoma conditions, sixteen Hashimoto's conditions and twenty five normal cases. The abnormal conditions on the Ultrasound images are drawn manually by the experienced radiologist and stored as a ‘ground truth’. The gray-level co-occurrence matrix and the gray-level run-length features of the ‘hand drawn’ region of interest (ROI) are extracted for each image. The features are then analyzed using the statistical unpaired two-tailed Student's t-test and classified based on the ‘ground truth’ data. The t-test shows a significant (p<0.001) differences between the groups. The texture features extracted from the Ultrasound images proves that the thyroid disorders such as Adenoma and Hashimoto's Thyroiditis can be distinguished from the normal using the highlighted texture features.