Publication

Abstract : Thyroid tumors and disorders are serious health concerns and the necessity for accurate diagnostic procedures are required. In this paper, we are going to build a model that predicts thyroid tumors using ultrasonic images and classifies them into three categories – benign, malignant, and normal thyroid. Using a large dataset of ultrasonic images, we employ deep learning techniques for Restnet101-based feature extraction and pattern recognition. Through examining and cross–validating, we improve the model’s performance and aim for high accuracy in categorizing between benign, malignant, and normal thyroid conditions and predicting the nodules. We implemented Resnet 101 which is a Convolutional neural network along with a training function to stop the iterations when there is no change in validation and training values to avoid overfitting and to reduce the time complexity. Our findings demonstrate that the model can efficiently aid doctors in making accurate diagnostic decisions, minimizing the need for invasive treatments, and improving patient outcomes. We Successfully deployed a model with an accuracy of 0.93 and Customised CNN having 4 layers with an accuracy of 0.873.