Η κλινική σημασία των προγνωστικών δεικτών για την έγκαιρη διάγνωση κακοήθων αλλοιώσεων του θυρεοειδούς αδένα σε υλικό παρακέντησης δια λεπτής βελόνης

Abstract

Purpose: The purpose of the present study is to investigate the capability of the combination of Learning Vector Quantizer (LVQ) Νeural Νetworks (NNs) in the discrimination of benign from malignant thyroid lesions.Patients and Method: The study was performed on Liquid Based Cytology (LBC) specimens taken by FNA and stained by Papanikolaou technique. From the cytological images, using a custom image analysis system, certain features, describing the size, shape and texture of approximately 100 nuclei per case, have been extracted. These features were used to classify each individual nucleus by an LVQ NN. In the sequel, the nucleus classification results for each case were used to classify each individual case by a second cascaded LVQ NN. The cases were distributed according to the histological diagnosis as follows: 165 cases were classified as goiter, 2 as nodular hyperplasias, 61 as Hashimoto thyroiditis, 3 as non specific thyroiditis, 21 as adenomatoid nodules, 3 as oxyphic adenomas and 4 cases as follicular adenomas. Αdditionally, 62 cases of papilary carcinoma, 9 of medullary carcinoma, 2 of anaplastic carcinoma and 3 of follicular carcinoma, were included.The data of about 50% of the cases from each class, were used for the training of two LVQ classifiers and the remaining data were used to test their performance. The proposed system was used to discriminate into the individual cellular level and subsequently into the individual patient level, between benign from malignant nuclei or cases.Results: The application of the LVQ NNs system allows successful discrimination between benign and malignant cell nuclei and lesions (overall accuracy 94,1% and 100% respectively).Conclusions: The results indicate that the use of neural networks combined with image morphometry may offer useful information on the potential of malignancy of thyroid lesions and could improve the diagnostic accuracy of FNA of the thyroid gland, especially in cases of follicular neoplasms classified as suspicious for malignancy and in cases of oncocytic tumors.