Thyroid Nodule Segmentation and Classification Using Deep Convolutional Neural Network and Rule-based Classifiers.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Thyroid cancer is globally prevalent, necessitating accurate diagnosis for effective treatment and reduced mortality.
• Ultrasound is a common tool for thyroid nodule monitoring, but faces challenges with over-diagnosis and assessment variability.
• Precise nodule boundary delineation for classification is time-consuming and prone to human error.

Purpose of the Study:

• To develop an automated, end-to-end pipeline for thyroid nodule segmentation and classification from ultrasound images.
• To enhance the accuracy and efficiency of thyroid nodule assessment, reducing diagnostic variability.
• To introduce a novel deep learning model for precise nodule boundary segmentation.

Main Methods:

• Proposed a residual dilated UNet (resDUnet) model for automatic segmentation of thyroid nodule boundaries.
• Integrated rule-based classifiers to categorize nodule composition and echogenicity from segmented images.
• Evaluated the segmentation performance on a dataset of 352 ultrasound images against radiologist-defined ground truth.

Main Results:

• The resDUnet model achieved a higher Dice score (82%) compared to standard UNet (81%) for nodule segmentation.
• The proposed method demonstrated robustness to variations in the user-specified region-of-interest.
• Minimal user interaction was required for the segmentation and classification pipeline.

Conclusions:

• The developed automated pipeline offers a more efficient and cost-effective approach to thyroid nodule assessment.
• The resDUnet model shows promise in improving the accuracy and consistency of thyroid nodule segmentation.
• This technology has the potential to reduce variability in clinical practice and improve thyroid cancer monitoring.