Artificial neural network model to distinguish follicular adenoma from follicular carcinoma on fine needle aspiration of thyroid. Issue 3 (20th December 2017)

• Record Type: Journal Article
• Authors: Savala, Rajiv; Dey, Pranab; Gupta, Nalini
• Is Part Of: Diagnostic cytopathology. Volume 46:Issue 3(2018)
• Page Start: 244
• Page End: 249
• Publication Date: 2017-12-20
• Subjects: artificial intelligence -- artificial neural network -- follicular adenoma -- follicular carcinoma -- thyroid; Cytodiagnosis -- Periodicals; Pathology, Cellular -- Periodicals; Cytology -- Periodicals; 616.07582
• DOI: 10.1002/dc.23880
• Languages: English
• ISSNs: 8755-1039

Abstract

Abstract : Background: To distinguish follicular adenoma (FA) and follicular carcinoma (FC) of thyroid in fine needle aspiration cytology (FNAC) is a challenging problem. Aims and objectives: In this article, we attempted to build an artificial neural network (ANN) model from the cytological and morphometric features of the FNAC smears of thyroid to distinguish FA from FC. Material and methods: The cytological features and morphometric analysis were done on the FNAC smears of histology proven cases of FA (26) and FC (31). The cytological features were analysed semi‐quantitatively by two independent observers (RS and PD). These data were used to make an ANN model to differentiate FA versus FC on FNAC material. The performance of this ANN model was assessed by analysing the confusion matrix and receiving operator curve. Result: There were 39 cases in training set, 9 cases each in validation and test sets. In the test group, ANN model successfully distinguished all cases (9/9) of FA and FC. The area under receiver operating curve was 1. Conclusion: The present ANN model is efficient to diagnose follicular adenoma and carcinoma cases on cytology smears without any error. In future, this ANN model will be able to diagnose follicular adenoma and carcinoma cases on thyroid aspirate. This study has immense potential in future. This is an open ended ANN model and more parameters and more cases can be included to make the model much stronger.