Thyroid Nodule Detection Using Deep Learning Strategies
Satuluri Naganjaneyulu, Rangisetty Sree Lakshmi, Shaik NagulMeeraBee, Medagam Sai Krishna Reddy · 2024
Detection of thyroid nodule is an important part in medical imaging because they occur more often and can be of any form from benign to malignant. If we detect the thyroid nodules in the beginning stages itself we can provide better treatment for the patient. The proposed model helps in thyroid nodules detection by using Convolutional Neural Network(CNN) and classifies thyroid nodules into functional, malignant and benign categories. First step involves loading and preprocessing the pictures dataset. These pictures contain various types of thyroid images collected from various medical imaging methods like ultrasound scanning reports. The images are shrunked, grayscaled and labelled based on file keyword names similar to real world diagnosis classification of medical photographs. The dataset is divided mainly into the training set and the testing set so that model is developed using the training data can be checked for it’s performance by extracting patterns from the testing data. During training, maintaining the consistency in pixel values between photos, data normalization keeps features of any image occupying the center stage. The CNN model is constructed using layers with an aim to identify complex patterns in image. Maxpooling layers are useful in extracting main features, where convolutional layers are used to find spatial patterns. Dense layers formed afterwards helps to develop the intricate representations which are required for a precise classification. After the model architecture is defined, the annotated photos are used to train the model architecture. By using the optimizer “Adam” and the “Categorical cross entropy loss function”, the model gets trained with the ten epochs which improve the models performance. It is very useful since it helps the doctors to quickly determine the precise type of thyroid nodules, therefore planning early diagnosis and therapy planning for the patient. This helps in speeding up the decision making by improving classification process, resulting prompt patient interventions. These type of automated methods also scale medical diagnostics which is helpful in providing services to large population round the globe by reaching out to the places with limited special healthcare.