Thyroid: proteins, genes, AI

Thyroid nodules are common, but the hard part is telling harmless lumps from cancer before surgery. Three new reports attack that problem with proteins, gene activity and artificial intelligence imaging. (cell.com) (aacrjournals.org) (springer.com) Doctors usually start with fine-needle aspiration, a biopsy that removes a small sample from a thyroid nodule. That test often works, but the ThyroProt study says some samples still land in an “indeterminate” zone that leaves patients and surgeons guessing. (cell.com) ThyroProt measures three proteins in the biopsy sample with targeted mass spectrometry, then combines that readout with BRAF V600E mutation status, age and sex. In a prospective, blinded, multicenter study, the team developed and validated the classifier on 837 fine-needle aspiration samples from 17 tertiary centers. (pmc.ncbi.nlm.nih.gov) In the prospective test set of 322 samples, ThyroProt reached an area under the curve of 0.94 and an overall accuracy of 90.7%, according to the paper in *Cell Reports Medicine*. The study focused on preoperative diagnosis, meaning the result is meant to guide decisions before a patient goes to the operating room. (pmc.ncbi.nlm.nih.gov) A separate line of work looks past diagnosis to prognosis: not just “is this cancer,” but “how likely is this cancer to come back.” Researchers at The University of Texas MD Anderson Cancer Center reported a papillary thyroid carcinoma gene signature called PRECISE in *Clinical Cancer Research* this week. (mdanderson.org) (aacrjournals.org) PRECISE was built from single-cell and single-nucleus RNA sequencing, which reads which genes tumor cells have switched on, and then tested across patient cohorts. The abstract says the discovery cohort included 109 patients at MD Anderson with a median follow-up of 14 years, plus sequencing from 11 tumors and four normal thyroid samples. (aacrjournals.org) A third approach tries to avoid the needle in the first place by getting more from ultrasound. A prospective 2025 study in *Endocrine* found multiparametric ultrasound outperformed artificial-intelligence-assisted B-mode ultrasound for solid thyroid nodules and reported potential to cut unnecessary biopsies compared with standard B-mode imaging. (springer.com) That ultrasound study enrolled 226 solid nodules from 189 patients, including 145 malignant and 81 benign lesions confirmed by pathology. The authors compared standard B-mode, artificial-intelligence-assisted ultrasound and multiparametric ultrasound, which combines several imaging signals instead of relying on one picture alone. (springer.com) Together, the three reports map the same clinical bottleneck from different angles: proteins for borderline biopsies, gene signatures for recurrence risk and richer imaging for biopsy triage. For patients with thyroid nodules, the near-term question is still the same one doctors face every day — who needs surgery, who needs closer follow-up and who can avoid another procedure. (cell.com) (aacrjournals.org) (springer.com)