A systematic review of machine learning based thyroid tumor characterisation using ultrasonographic images

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare

Background:

• Ultrasonography (US) is a safe, accessible tool for thyroid tumor screening.
• Image artifacts and speckle noise in US hinder early detection of thyroid abnormalities.
• Continuous research aims to enhance US image analysis for improved diagnostic accuracy.

Purpose of the Study:

• To comprehensively review computer-aided diagnosis (CAD) systems for thyroid tumor US (TTUS) images.
• To identify and analyze key components of TTUS CAD systems, including datasets, algorithms, and classification methods.
• To provide a roadmap for future research in this field.

Main Methods:

• Extensive literature review of CAD systems for TTUS image classification.
• Analysis of various components: datasets, despeckling algorithms, segmentation algorithms, feature extraction/selection, assessment parameters, and classification algorithms.
• Synthesis of achievements and challenges in the field.

Main Results:

• Detailed overview of diverse TTUS datasets and their characteristics.
• Evaluation of various image processing techniques (despeckling, segmentation) applied to TTUS images.
• Summary of feature extraction, selection, and classification methodologies used in TTUS CAD systems.

Conclusions:

• Significant progress has been made in developing CAD systems for thyroid US image analysis.
• Challenges remain in overcoming image artifacts and improving diagnostic performance.
• The study provides a roadmap to guide future research and development in thyroid tumor ultrasonography analysis.