Deep Learning-based Image Analysis Model for Diagnosing Thyroid Carcinoma in Fine Needle Aspiration Cytology (FNAC) Images

In this work, the diagnostic accuracy of an automated diagnosis system is evaluated using two pre-trained convolutional neural network models, namely AlexNet and VGG16.The diagnosis system is used to identify the cancerous and normal thyroid cells in Fine Needle Aspiration Cytology (FNAC) photographs.The proposed Alexnet and VGG16 models are implemented using deep learning based Transfer Learning (TL) to process multi-stained FNAC images.Initially, the image patches are derived from the cytology images based on the thyroid cell population.These patches are fed to the 8-deep layered AlexNet and 16-deep layered VGG16 as inputs and they are passed through multiple convolution layers, max pooling and dense layers.Through optimal implementation and testing of the models, the AlexNet model achieves a diagnostic accuracy of 92.5% whereas the VGG16 model results a diagnostic accuracy of 96.66% and sensitivity and specificity of 98.75% and 92.5% respectively.