Kiwi Toku Eyes’ artificial intelligence (AI) tool Theia, achieved a high accuracy (94-95%) rate for detecting referable diabetic retinopathy (DR) in a newly published, multi-ethnic study.
Designed to assist in clinical decision‐making and tailored to the needs of the New Zealand national DR screening programme, Theia is the only AI tool in the world that has captured Māori and Pasifika in its training and validation, said co-developer, Toku Eyes’ co-founder Dr Ehsan Vaghefi from the School of Optometry and Vision Science (SOVS) at Auckland University.
The Toku Eyes’ team developed its AI system using routinely collected retinal screening datasets from Auckland District Health Board (ADHB) and the Counties Manukau District Health Board (CMDHB) from January 2009 to December 2018. Images were labelled as non‐sight‐threatening, potentially referable or sight‐threatening for New Zealand implementation, or as referable (potentially referable + sight‐threatening)/non‐referable (non‐sight‐threatening) for global comparison.
Screening data from 32,354 diabetic patients (63,843 including multiple visits) were analysed, of which 95–97%, 0.9–2.4% and 1.1–3.1% were categorised as non‐sight‐threatening, potentially referable and sight‐threatening, respectively. Using the referable/non‐referable categories, Theia achieved an overall sensitivity of 94% in the ADHB dataset and 95% in the CMDHB set, while preserving overall negative predictive value of 99.5%.
Theia was designed to have extremely high sensitivity and negative predictive value, said Dr Vaghefi. “The aim from the beginning was to ‘miss no disease’.” Although the high sensitivity comes at the price of a higher number of false positives, ie. lower specificity, Vaghefi said the team is happy with this design. “A small number of additional false positives will not impose big challenges to the screening services, since we have already reduced their workload by 50%.”
