Progetti di ricerca

Nodular thyroid disease affects approximately 68% of the population, predominantly women, with up to 35% of nodules hiding thyroid tumors, where papillary thyroid carcinoma (PTC) is the most frequent in young adults. The identification of these cases is based on… Leggi tutto ultrasound (US) and cytological evaluation by fine needle aspiration (FNA), but still presents gray areas of diagnostic uncertainty. In this context, the category of non-invasive follicular tumors with papillary-like nuclear features (NIFTP), recently introduced as a less aggressive alternative to PTC, imposes an important update of our diagnostic schemes. Previous experiences had already tried to identify characteristics capable of distinguishing NIFTP from PTC, with the aim of improving the surgical management of these cases to avoid diagnostic thyroidectomies ("thyroid carnage"), but this has not yet significantly contributed to the routine histopathological diagnostics. The introduction of Artificial Intelligence (AI) and computational pathology (CP) has recently demonstrated the ability to discriminate NIFTP from benign nuclei and PTC on histology (NUTSHELL: NUclei from Thyroid tumors Segmentation to Highlight Encapsulated Low-malignant Lesions) , enabling the definition of human interpretable features (HIF) and the development of a classification system to assist the pathologist on whole slide images (WSI) after surgery. The aim of the CLARIFAI project is to improve pre-surgical FNA classification of thyroid lesions through a nuclear classifier and HIF introduced with AI and digital pathology. To achieve this goal, the next steps are to apply this computational tool on preoperative thyroid cytology samples to aid pathologists in distinguishing NIFTP cases from PTC cases. For this two different approaches will be adopted: - extension of the capabilities of the already consolidated histology-based classifier NUTSHELL on cytological samples; - application of multiple instance learning (MIL) on labeled cytology samples