Drop the shortcuts: image augmentation improves fairness and decreases AI detection of race and other demographics from medical images

BACKGROUND

It has been shown that AI models can learn race on medical images, leading to algorithmic bias. Our aim in this study was to enhance the fairness of medical image models by eliminating bias related to race, age, and sex. We hypothesise models may be learning demographics via shortcut learning and combat this using image augmentation.

METHODS

This study included 44,953 patients who identified as Asian, Black, or White (mean age, 60.68 years ±18.21; 23,499 women) for a total of 194,359 chest X-rays (CXRs) from MIMIC-CXR database. The included CheXpert images comprised 45,095 patients (mean age 63.10 years ±18.14; 20,437 women) for a total of 134,300 CXRs were used for external validation. We also collected 1195 3D brain magnetic resonance imaging (MRI) data from the ADNI database, which included 273 participants with an average age of 76.97 years ±14.22, and 142 females. DL models were trained on either non-augmented or augmented images and assessed using disparity metrics. The features learned by the models were analysed using task transfer experiments and model visualisation techniques.

FINDINGS

In the detection of radiological findings, training a model using augmented CXR images was shown to reduce disparities in error rate among racial groups (-5.45%), age groups (-13.94%), and sex (-22.22%). For AD detection, the model trained with augmented MRI images was shown 53.11% and 31.01% reduction of disparities in error rate among age and sex groups, respectively. Image augmentation led to a reduction in the model's ability to identify demographic attributes and resulted in the model trained for clinical purposes incorporating fewer demographic features.

INTERPRETATION

The model trained using the augmented images was less likely to be influenced by demographic information in detecting image labels. These results demonstrate that the proposed augmentation scheme could enhance the fairness of interpretations by DL models when dealing with data from patients with different demographic backgrounds.

FUNDING

National Science and Technology Council (Taiwan), National Institutes of Health.

Are spray drift losses to agricultural roads more important for surface water contamination than direct drift to surface waters?

Spray drift is considered a major pesticide transport pathway to surface waters. Current research and legislation usually only considers direct spray drift. However, also spray drift on roads and subsequent wash-off to surface waters was identified as a possible transport pathway. Hydraulic shortcuts (storm drainage inlets, channel drains, ditches) have been shown to connect roads to surface waters, thus increasing the risk of drift wash-off to surface waters. However, the importance of this pathway has never been assessed on larger scales. To address this knowledge gap, we studied 26 agricultural catchments with a predominance of arable cropping (n = 17) and vineyards (n = 9). In these study sites, we assessed the occurrence of shortcuts by field mapping. Afterwards, we modelled the areas of roads drained to surface waters using a high-resolution digital elevation model (0.5 m resolution) and a multiple flow algorithm. Finally, we modelled drift deposition to drained roads and surface waters using a spatially explicit, georeferenced spray drift model. Our results show that for most sites, the drift to drained roads is much larger than the direct drift to surface waters. In arable land sites, drift to roads exceeds the direct drift by a factor of 4.5 to 18, and in vineyard sites by 35 to 140. In arable land sites, drift to drained roads is rather small (0.0015% to 0.0049% of applied amount) compared to typical total pesticide losses to surface waters. However, substantial drift to drained roads in vineyard sites was found (0.063% to 0.20% of applied amount). Current literature suggests that major fractions of the drift deposited on roads can be washed off during rain events, especially for pesticides with low soil adsorption coefficients. For such pesticides and particularly in vineyards, spray drift wash-off from drained roads is therefore expected to be a major transport pathway to surface waters.