Ultrasonography Diagnostics Using Gaussian Mixture Model
Daniel Smutek · 2003
ture modeling Abstract: The problem of diagnostics of diseases of thyroid gland is addressed, namely Hashimoto’s Lymphotitic thyroiditis, which is one of most usual thyroid gland diseases. As opposed to expensive diagnosis using magnetic resonance or by painful invasive methods, we base our method on using ultrasound images. Sonographic imaging of thyroid gland is done rou-tinely and these images can be used to find new cases which can help to start treatment as soon as possible. It is hard for physician, to make thyroid gland diagnostics based on ultrasound image only. However, we have found out (see [2, 3]), that Hashimoto’s Lymphotitic thyroiditis changes the pattern of the ultrasound scan. For diagnostics, we use the statistics method. Database of ultrasound images with known diagnosis is used for learning. Pattern features are calculated and Gaussian mixture model in the space of pattern features is used. In diagnostics stage, the sonogram of person with unknown diagnostics is taken and the most probable diagnosis is inferred using Bayesian inference with on Gaussian clusters of pattern features. Hashimoto’s lymphocytic thyroiditis is ideally suited