Effects of Hybrid Contrast Enhancement and Bilateral Filtering for Enhancing Ultrasound Thyroid Cancer Images Classification

Artificial intelligence (AI), in particular deep learning algorithms, has made great strides in image classification tasks, enabling the autonomous evaluation of intricate medical images. This is especially important when using ultrasonography, to diagnose thyroid cancer. AI can lessen radiologists' burden, improve image processing efficiency, and assist them in distinguishing between benign and malignant nodules. This method has a lot of potential to increase the accuracy of thyroid ailments ultrasound diagnosis. In this study Random Forest and Support Vector Machine were employed to classify cancerous and non-cancerous thyroid images, by extracting deep learning features of five Deep Learning (DL) models namely ResNet50, ResNet101, VGG16, VGG19, and MobileNet. The performance of the models was compared. To improve the performance of the models the images were filtered using Bilateral Filter (BF) and Enhanced using Contrast Enhancement (CEH) technique. Also, to further improve the performance of the models, the models were trained with images to which the CEH and BL were applied. The best-performing model turns out to be CE+BF+ResNet50-SVM with an accuracy of $\mathbf{9 7 . 0 2 \%}$, sensitivity of $\mathbf{9 7 \%}$, specificity of $\mathbf{9 6 . 0 3 \%}$, F1-Score of $\mathbf{9 6 . 4 9 \%}$, AUC of $\mathbf{9 7 . 0 2 \%}$ and precision of $\mathbf{9 5 . 9 8 \%}$.