Proposed Topic (Most preferred): :
Research and Innovations (new projects / technology / innovations / service models)
Proposed Topic (Second preferred): :
HA Young Investigators Session (Projects to be presented by HA staff who had joined HA for 10 years or less)
Authors (including presenting author) :
Fenn D(1), Fok WSE(1), Fung PYE(1), Mak WS(1), Cheung WP (1), Sit WI(1), Ng WK (1), Chan CYG(1), Lo LW (1), Wong KML(1), Kwok KM(1)
Affiliation :
(1) Department of Diagnostic and Interventional Radiology, Kwong Wah Hospital
Introduction :
Breast cancer is the most common cancer among females in Hong Kong. Performance of AI varies between different studies. Some studies suggested the potential role of AI in population-based breast cancer. One of which demonstrated that AI can improve the performance of radiologists without prolonging the workflow. Local evidence of AI-assisted interpretation of mammograms in a screening context is lacking.
Objectives :
(1) To determine whether AI improves the performance of human readers (2) To determine the experience of human readers with AI assistance
Methodology :
The AI algorithm used was a commercially available product (Lunit INSIGHT MMG, version 1.1.7.3). The Hong Kong Personal Performance in Mammographic Screening Scheme (HKPERFORMS) test set was developed to assess reader performance. The study utilised the HKPERFORMS test set, which was based on an established test set used in the UK for 30 years. Human readers will evaluate the HKPERFORMS test set both with and without AI assistance. Their performance and experience will be assessed.
Result & Outcome :
Results
The HKPERFORMS test set includes 40 abnormal and 80 normal sets of mammogram(s). The ground truth of abnormal cases were established with pathologically proven pathologies and normal or benign cases were established when normal or benign findings were stable for at least 3 years. The outcome of this study is still pending.
Conclusions
This is the first study evaluating the performance and experience of human readers with and without AI assistance using a mammogram test set derived from the screening population in Hong Kong. This study may shed light on the possibility of incorporating of AI as an assistant in a routine mammographic interpretation workflow, particularly in the screening population.
