The thyroid nodule is one of the endocrine issues caused by an irregular cell development in the thyroid which might be in one of the two categories benign or malignant.
The objectives of the present study are to design a new image processing and analysis methods in ultrasound thyroid images to screen nodule malignancy. The research procedure comprised two main concepts to achieve this goal. Firstly, to perform the automated diagnosis of thyroid nodule malignancy using texture information derived from Gary level co-occurrence, run-length and statistical difference, statistical tools from both of spatial and Two Dimensional Wavelet(2D-DWT), two levels (eight bands) decomposition domains. Secondly, a hybrid system that adopt the successful features from these two domains throughout ANOVA feature selection system. Hence, 100-thyroid nodule images out of 300- ultrasound images spliced equally between the two categories (50 Benign and 50 Malignant) have been selected. The Textural features are made up of (19) spatial domain statistical features derived from texture analysis tools GLCM (Gray Level Co-occurrence Matrix), GLRLM (Gray Level Run Length Matrix), and Statistical Differences (SD). After quantization of wavelet coefficients, the same features were generated from four bands (LL, HL, LH, HH) at two levels of decompositions. Eventually out of 171 features retrieved, 51 succeed features were selected from both of spatial domain and hybrid features, considering the P-values less than or equal to 0.001.
The first proposed system shows thyroid nodules malignancy with an average accuracy of about 71% using SVM classifier for all features derived from spatial domain while the average accuracy is increased to 98% in case of hybrid feature. For the other proposed classifier (DT), the average accuracy in case of spatial domain based features was 73% whereas the average accuracy of the hybrid features system was 95%. Finally, based on (KNN) classifier, the average accuracy in case of spatial domain was 69% and in case of hybrid feature was 88%. Hence the proposed hybrid system outperform classical textural features system based only on spatial domain and can efficiently be used for screening thyroid nodules malignancy. For the given dataset, the SVM second scenario classifier shows the best classification Accuracy of 98% Sensitive (Recall) of 100%, Specificity (Kappa) of 96% and Precision of 94 % using 51 ranked features.
In the second proposed system a novel neutrosophic method has been used to enhance the medical images of thyroid nodule to facilitate the automatic selection of a seed point in thyroid nodule. The proposed system consists of three components: neutrosophic image enhancement, speckle reduction and automatic seed selection algorithm extracted from the center of candidate block in ultrasound thyroid images based on the principle that most of its Higher Order Spectra Entropies (HOSE) from Radon Transform (RT) at different angles are within the range between average and maximum entropies. Eventually the region growing image segmentation is applied with the constant threshold.
The performance of proposed automatic segmentation method has been compared with other methods in terms of calculating, True Positive (TP) value (96.44 ± 3.01%), False Positive (FP) value (3.55 ± 1.45%), Dice Coefficient (DC) value (92.24 ± 6.47%), Similarity Index (SI) (80.57 ± 1.06%), and Hausdroff Distance (HD) (0.42 ± 0.24 pixels). The proposed system can be considered as an added value to the malignancy diagnosis in thyroid nodule by an endocrinologist.