TY  - GEN
SP  - 78
AV  - restricted
Y1  - 2024/10/13/
EP  - 87
TI  - Exploring Fairness in State-of-the-Art Pulmonary Nodule Detection Algorithms
N1  - This version is the author accepted manuscript. For information on re-use, please refer to the publisher?s terms and conditions.
PB  - Springer, Cham
UR  - http://dx.doi.org/10.1007/978-3-031-72787-0_8
SN  - 0302-9743
N2  - Lung cancer is the leading cause of cancer mortality worldwide. Asymptomatic in its early stages, it is disproportionately detected when the disease is advanced. Resource constraints have resulted in increasing reliance on computer-aided detection (CADe) systems to assist with scan evaluation. The datasets used to train these algorithms are often unbalanced in their representation of protected groups e.g. sex and ethnicity. This project investigates whether there are performance disparities in detecting clinically relevant nodules across under-represented groups in selected, state-of-the-art nodule detection algorithms trained on data from a screening program in the UK. Our analysis revealed that overall, the algorithms demonstrate equitable performance across various demographic groups. However, their performance varies strongly across nodule characteristics (size and type) in line with their prevalence in the training set. To ensure continued equitable performance, algorithms should not only consider demographic but also nodule attributes representativeness in their training.
ID  - discovery10200673
A1  - McCabe, J
A1  - Cheng, D
A1  - Bhamani, A
A1  - Mullin, M
A1  - Patrick, T
A1  - Nair, A
A1  - Janes, SM
A1  - Sudre, CH
A1  - Jacob, J
T3  - Lecture Notes in Computer Science, volume 15198
KW  - Nodule Detection Algorithms
KW  -  Fairness in AI
KW  -  Lung Cancer
Screening
ER  -