Thyroid cancer detection and classification using spectral imaging and artificial intelligence.

Thyroid cancer is the most prevalent endocrine cancer, with a steadily rising incidence. Its diagnosis involves microscopic examination of tissue specimens, a process that can lead to misclassification with critical prognostic consequences. Numerous artificial intelligence techniques have been proposed for thyroid cancer detection, however, none have proven clinically relevant. We present an accurate diagnostic approach based on a newly developed spectral imaging system that rapidly measures the visible spectrum at each point on routinely prepared hematoxylin and eosin-stained tissue sections. These spectral images are analyzed with machine learning algorithms, classifying each nucleus while preserving interpretability for experts. The integration of spectral imaging and artificial intelligence enables precise, robust identification of normal and tumor cells, offering a straightforward, powerful approach for thyroid cancer assessment. By utilizing routinely stained tissue specimens and targeting specific pathological features, our method provides a tool to support pathologists, facilitating accurate and timely evaluations of thyroid cancer.