What is it about?

In our work, we report the application of a custom fast-scanning wide-field SHG microscope capable of imaging sub-centimeter regions with sub-micrometer resolution in combination with unsupervised machine learning (ML) algorithms for the analysis of papillary thyroid carcinoma (PTC). Our paper aims at detecting invaded regions in the collagen capsule surrounding the carcinoma. Unsupervised ML of SHG scans enabled revealing specific features within multiple large-scale samples based only on collagen texture and intensity features. The discovery of meaningful subgroups in the datasets allowed us to qualitatively and quantitatively demonstrate the textural heterogeneity of the PTC capsule and assign feature sets to specific changes in collagen networks, preconditions for invasion, and invasion itself.

Featured Image

Why is it important?

We demonstrate the diagnostic potential of ML-enhanced fast-scanning wide-field SHG microscopy and hypothesize that the proposed approach may help to reveal poorly distinguishable invasions and highlight areas of the PTC capsule that require closer and more careful examination. Timely detection of invasive areas is very important for correct diagnosis and choice of treatment.

Perspectives

The proposed approach can significantly contribute to revealing non-obvious, sometimes accidently missed microinvasions in PTC nodules, as shown in the study, and greatly accelerate and simplify the development of reliable methods for fully automated ML diagnosis that can be integrated into clinical practice.

Dr. Lena N Golubewa
State research institute Center for Physical Sciences and Technology (Valstybinis mokslinių tyrimų institutas Fizinių ir technologijos mokslų centras (FTMC))

Read the Original

This page is a summary of: Machine learning-based diagnostics of capsular invasion in thyroid nodules with wide-field second harmonic generation microscopy, Computerized Medical Imaging and Graphics, October 2024, Elsevier,
DOI: 10.1016/j.compmedimag.2024.102440.
You can read the full text:

Read

Resources

Contributors

The following have contributed to this page