A machine-learning algorithm that analyzes thyroid-nodule ultrasound images correctly identifies likely cancerous nodules 97% of the time - which is on par with thyroid biopsy with fine-needle aspiration, according to research presented at the Endocrine Society annual meeting.