Machine learning-based diagnostics of capsular invasion in thyroid nodules with wide-field second harmonic generation...

Area of Science:

• Biomedical Optics
• Computational Pathology
• Cancer Research

Background:

• Papillary thyroid carcinoma (PTC) is a common thyroid cancer.
• The collagen capsule surrounding PTC nodules is a barrier to invasion.
• Detecting capsular invasion is vital for diagnosis and treatment planning.

Purpose of the Study:

• To quantitatively characterize collagen structure in PTC capsules using SHG microscopy and machine learning.
• To differentiate between intact, invaded, and potentially invaded capsule regions.
• To enhance diagnostic performance for encapsulated PTC.

Main Methods:

• Wide-field second harmonic generation (SHG) microscopy.
• Texture analysis.
• Unsupervised machine learning (k-means clustering).

Main Results:

• Collagen capsules are heterogeneous with distinct segments identified by ML parameters.
• Invasive sites show fragmented, curled collagen fibers.
• ML identified potential micro-invasion areas missed by initial histopathology.

Conclusions:

• Unsupervised ML with SHG microscopy can reliably complement PTC diagnosis.
• This approach improves diagnostic reliability, simplifies decision-making, and reduces errors.
• Automated ML methods can accelerate the development of fully automated diagnostic tools for clinical integration.