Quantitative data in relation to diagnostic information were obtained from scintiscan matrices by digital computer. The data, from scintillation scanner through punched tape were smoothed by the moving average method, and the perimeter of thyroid was decided from histogram of smoothed data, and then counts were normalized from 0 to 10. Thyroid was divided into right and left lobes from maximum and minimum points of curvature of the perimeter.
About each lobe individually, features such as uptake of radioactive iodine, quantification of 131I distribution, difference of 1311 distribution between right and left lobes, area of thyroid, ratio of the longest and the shortest diameters, and perimeter irregularity quantitated by circle fitting to the perimeter were extracted and by using decision tree combined with linear discriminant function, thyroid scintigrams were classified into eight categories. These categories are normal, hyperthyroidism, hot nodule, small defect in lobe, large defect of unilateral lobe, large defect of both lobes, complete defect of unilateral lobe and low uptake.
By this method, about 85% of pathological information on thyroid was correctly abstracted.