Logistic regression in the analysis of image cytometric data of fine-needle aspirated cells of breast cancer patients-a comparison with artificial neural networks

Image cytometry data of aspirated tumour cells from 102 patients with breast cancer were analysed and used as prognostic markers in an attempt to predict involvement of axillary lymph nodes and histological grade using logistic regression. Prediction was 70% for both nodal status and histological analyses. The outcome of this study is compared to an earlier study using the same cytological information to obtain prediction using a neural approach. Using artificial neural networks, prediction accuracy was 87% and 82% for nodal status and histological assessment, respectively. This study also attempts to identify the impact of individual prognostic factors. The statistical approach identified S-phase fraction and DNA-ploidy as the most important prediction markers for nodal status and histological assessment analyses. A comparison was made between these two quantitative techniques.