A Pattern Recognition System for the Comparative Evaluation of Physician's Subjective Assessment Versus Quantitative Nuclear Features in Grading Urine Bladder Tumors

P. Spyridonos, D. Cavouras, P. Ravazoula, and G. Nikiforidis (Greece)

Keywords

Classification; Grading of bladder carcinoma; nuclear features

Abstract

A pattern recognition (PR) system was developed to compare the diagnostic effectiveness of subjective criteria employed by physicians against quantitative measures used in computer based texture analysis for grading urine bladder tumours. Ninety-two cases were classified according to WHO grading system by 4 experienced pathologists in three classes: grade I, grade II and grade III. Each case was represented by eight histological features routinely employed by pathologists. Features were quantified for use by the PR system following a standard procedure. Additionally, in each case images from tissue samples were processed and 36 morphological and textural nuclear features were extracted. The PR-system employed a neural network classifier for the automatic segregation of tumors’ grade. Classifier accuracy employing subjectively evaluated features was 65.6%, 69.4%, 82.7% for grades I, II, III respectively. Using quantitative measures of nuclear features in the PR-system classification performance improved to 63.52 %, 81.62%, and 89.6% for grades I, II, III respectively. In conclusion, textural and morphologic nuclear features may incorporate additional valuable information to the physician for grading tumors. Also, when these features are employed in a PR system they may provide a 2nd objective opinion to the pathologist in the differentiation of urine bladder tumors.

Important Links:



Go Back