Advanced thyroid nodule detection using ultrasonography image analysis and bilateral mean clustering strategies

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Thyroid nodule diagnosis relies on imaging, but accurately assessing severity, especially concerning potential malignancy, remains challenging.
• Calcium flecks within thyroid nodules are recognized indicators of severity, potentially correlating with thyroid cancer.
• Existing diagnostic methods may lack the precision needed to quantify calcium fleck presence and its implications for malignancy.

Purpose of the Study:

• To develop an advanced diagnostic method for thyroid nodules using ultrasonography images.
• To enhance the accuracy of thyroid nodule segmentation and severity assessment by precisely analyzing calcium flecks.
• To introduce and evaluate a novel deep convolutional neural network (DCNN) incorporating a new clustering strategy for improved diagnostic performance.

Main Methods:

• A novel Bilateral Mean Clustering Strategy (Bi-MCS) was developed, integrating Fuzzy C-means and K-means clustering.
• Bi-MCS was employed to improve color sense-based segmentation accuracy by precisely identifying thyroid nodule edges through analysis of pixel intensity variations related to calcium flecks.
• The Bi-MCS technique was integrated into a modified ResNet101 architecture, creating the Bi-ResNet101 DCNN for analyzing grayscale intensity profiles and calcium fleck density.

Main Results:

• The Bi-ResNet101 model demonstrated superior performance in computing calcium fleck density compared to standard ResNet models (Resnet18, Resnet50, Resnet101).
• Experimental analysis confirmed the enhanced capability of Bi-ResNet101 in accurately classifying different stages of thyroid nodules based on calcium fleck analysis.
• The proposed method achieved higher precision in identifying nodule edges, crucial for accurate severity determination.

Conclusions:

• The developed Bi-ResNet101 model, utilizing the Bi-MCS technique, offers a significant advancement in the non-invasive diagnosis of thyroid nodules.
• Accurate quantification of calcium flecks in ultrasonography images via this method can lead to more precise thyroid nodule severity classification and potentially earlier detection of thyroid cancer.
• This approach holds promise for improving diagnostic accuracy and patient outcomes in thyroid nodule management.