APPLICATION OF MULTIVARIATE, FUZZY SET AND NEURAL-NETWORK ANALYSIS INQUANTITATIVE CYTOLOGICAL EXAMINATIONS

Multivariate statistical methods have been used in several studies to increase the diagnostic reliability of TV image analyser systems. In r ecent years some algorithms for decision support (fuzzy logic) and for pattern recognition (neural nets), both non-linear, were developed. T his paper reports on preliminary results obtained with these methods i n quantitative cytology and compares them to the traditional classifie rs. A total of 21 normal, 15 dysplastic and 23 malignant, gastric impr int smears were Feulgen stained and analysed on a Leitz Miamed DNA cyt ophotometer system. Mean DNA content, the 2c deviation index (2cDI), 5 c exceeding rate (5cER), G1, S, G2 phase fraction ratios, cell nucleus area and form factor were determined. Diagnostic accuracy of the disc riminant analysis was 96% for the malignant cases, 87% for dysplasias and 81% for normal cases. Cluster analysis gave no significant result. Our diagnostic system utilizing fuzzy logic has made the diagnostic b orders adjustable and reliable. The back-propagation neural net correc tly classifed the normal and malignant cases (100%) and all but one of the dysplasias (98%). The non-linear mathematical methods improved th e reliability of the diagnostic system. These new algorithms gave resu lts comparable to traditional classifiers. The application of these me thods to clinical samples is encouraging.