Background
Thyroid cancer is the 9th commonest cancer and the 4th among women. It is diagnosed by ultrasound and biopsy of radiologically suspicious nodules. Biopsy results are often inconclusive leading many patients to undergo thyroidectomy to obtain a definitive diagnosis.
The aim is design AI algorithms to interpret ultrasound images combined with clinical risk factors to better determine cancer risk. This may mitigate diagnostic thyroid surgery to make a cancer diagnosis. This may have an additional benefit of streamlining the ultrasound service for the very large number of patients possessing thyroid nodules.
Impact of ENT UK Foundation funding
The funding has been helpful in a portion of the de-identification which is an essential step in the process of data processing in any AI imaging work.
Current Status
Progress is on track.
We are undergoing extensive data extraction and processing of the thyroid ultrasound images. This is a laborious process, requiring extractions, anonymization, identifying a region of interest – the thyroid nodule in the image.
We have the basic framework designed for 2 different artificial intelligence tools that will require further editing and adjusting to maximise diagnostic accuracy of the algorithm once the full dataset has been processed.
Preliminary work demonstrates a diagnostic “area under the curve” (AUC) of 0.74, when compared to the literature for radiologists who demonstrate an average AUC of 0.83. Whilst not as good as radiologists this accuracy will only improve as we expand numbers and develop. We have been awarded the prestigious British Association of Endocrine and Thyroid Surgeons Research Grant based on the preliminary unpublished work showing that an artificial intelligence algorithm can be successfully developed using UK based images.
View more