Artificial Intelligence in Thyroid Nodules

  • Research type

    Research Study

  • Full title

    The use of artificial intelligence for risk stratification and prognostication in thyroid nodule ultrasound

  • IRAS ID

    297032

  • Contact name

    Zaid Awad

  • Contact email

    zawad@nhs.net

  • Sponsor organisation

    Imperial College Healthcare NHS Trust

  • Duration of Study in the UK

    1 years, 7 months, 1 days

  • Research summary

    Thyroid cancer is the most common endocrine cancer, in the UK. Thyroid nodules may have benign or malignant pathology. Thyroid nodules are very common with approximately 5% of women having palpable lumps with many more picked up on ultrasound scanning. The best investigation to rule out cancer is ultrasound. Unfortunately, the results of these tests are frequently “indeterminate”. Ultimately, this results in diagnostic surgery which carries risks such as bleeding, hoarseness and scarring. 75% of these nodules end up being diagnosed as benign after removal. More research is needed to explore new techniques that discriminate between cancer and non-cancer in thyroid lumps. A trained artificial intelligence program may improve diagnostic accuracy of imaging and avoid unnecessary operations, thus reducing surgical complications, patient anxiety and costs to the NHS. Furthermore, ultrasound interpretation is operator dependent and hence there is potential human error and variability in its classification. Recommendations of research to develop more accurate techniques in diagnosing thyroid nodules have been made by the British Thyroid Association. Currently there is no software being routinely used in patients with thyroid nodules in the UK. AI diagnosis decision support system to assess thyroid nodules has the potential to reduce the need for radiologists to interpret images (undertaken by others) and thereby reducing variability, standardising assessment, and making better use of radiologists’ time. Our aim is to design an AI based algorithm to support radiologist and radiographers by automatically differentiating between benign and malignant thyroid nodules.

  • REC name

    N/A

  • REC reference

    N/A