APPLICATION OF THE LEARNING VECTOR QUANTIZER TO THE CLASSIFICATION OFBREAST-LESIONS

OBJECTIVE: To investigate the potential of the learning vector quantiz ation (LVQ) neural network for the discrimination of benign from malig nant breast lesions. STUDY DESIGN: Using a custom image analysis syste m on Giemsa-stained smears, 25 parameters describing the size, shape a nd texture of the cell nucleus were measured. Three thousand nuclei fr om a total of 9,356 were selected as a training set for the neural net work, and the whole data set was used for testing. An additional 238 c ells from 16 cases without final cytologic diagnoses were evaluated by the system. The total number of cells (9,594) was collected from 100 patients (68 carcinomas and 32 benign lesions). RESULTS: Cytologic exa mination of the cases gave two false positive and two false negative r esults. However, in eight cases of ductal breast carcinoma and in eigh t cases of benign lesions, histologic confirmation was necessary in or der to confirm the cytologic diagnosis. Application of the LVQ permite d correct classification of 87.41% of the cells. Classification at the patient level by using a hypothesis test for proportion with a hypoth esis value equal to 50% permitted the correct diagnosis in 98% of pati ents. CONCLUSION: These results indicate that the use of neural networ ks combined with image morphometry and statistical techniques may offe r useful information about the potential for malignancy, improving the diagnostic accuracy of fine needle aspiration of breast lesions.