Texture and shape analysis of diffusion-weighted imaging for thyroid nodules classification using machine learning.

Area of Science:

• Radiology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Thyroid nodule classification relies on accurate diagnostic methods.
• Noninvasive techniques are crucial for improving diagnostic accuracy.
• Magnetic Resonance Imaging (MRI) offers versatile imaging capabilities.

Purpose of the Study:

• To evaluate if integrating diffusion-weighted imaging (DWI) functional features with T2-weighted MRI shape, texture, and volumetric features can noninvasively enhance thyroid nodule classification accuracy.
• To assess the diagnostic performance of a combined imaging approach for thyroid nodules.

Main Methods:

• Retrospective analysis of 55 patients with pathologically proven thyroid nodules.
• Acquisition of T2-weighted and diffusion-weighted MRI scans.
• Extraction of apparent diffusion coefficient (ADC) maps, nodule morphology (spherical harmonics, volume), and texture features (histogram statistics).
• Development of an artificial neural network (NN) fusion system to integrate functional, morphological, and texture features.
• Validation using leave-one-subject-out (LOSO) cross-validation.

Main Results:

• Functional, morphological, and texture imaging features were successfully extracted from 55 patients.
• The computer-aided diagnosis (CAD) system's accuracy improved with feature integration.
• The fusion system achieved high sensitivity, specificity, and accuracy (specific values omitted due to placeholder math symbols).

Conclusions:

• Integrating functional (DWI) with structural (T2-weighted MRI) imaging features shows promise for improved thyroid nodule identification.
• Machine learning approaches, particularly neural networks, are effective in fusing multimodal imaging data for enhanced diagnostic accuracy.
• This combined approach holds potential for noninvasive diagnosis of thyroid nodule malignancy.