Mr. Andrew Miller Profile

Andrew Miller

SPIE Involvement:

Author

Publications (1)

Proceedings Article | 6 March 2023 Presentation + Paper

Evaluating color correction algorithms for automated interpretation of urinalysis dipsticks with low-cost image sensors

Wenbo Wang, James Stafford, Andrew Miller, Rose Buchmann, Ethan Spencer, Angela Michelle San Juan, Jamie Purcell, Matthew Keller

Proceedings Volume 12369, 123690B (2023) https://doi.org/10.1117/12.2653972

KEYWORDS: Light sources and illumination, Imaging systems, Cameras, Glucose, Spectrophotometry, Color, Gamma correction, Proteins, Data analysis, Color difference

Read Abstract +

Some medical diagnostic tools, such as urinalysis dipsticks, rely on reading colors accurately for making clinical decisions. Reading color visually can be subject to 1) perceptual differences (inter-observer), 2) environmental factors such as illumination, and 3) target coloring (metamerism), especially among users with limited training and experience. Mobile phone cameras and compact camera modules offer potential low-cost platforms for automated, objective color readouts. However, image colors are a function of camera sensor and illumination characteristics. To restore color fidelity, color correction techniques must be applied to account for systematic deviation. This work aims to provide a quantitative assessment of color correction techniques and reduce variability in color interpretation for urinalysis dipstick results using a low-cost imager. Three color correction methods – linear, polynomial, and root-polynomial regression – were compared for performance in color difference reduction. A standard color checker card was used as reference to compute color correction matrices. A custom imaging system with a low-cost camera module was developed to capture images under controlled illumination. Reference values of the color checker card were obtained with a CM-26d handheld spectrophotometer. The CIE2000 ∆E was used to quantify the color difference between the camera image and the spectrometer to evaluate 3 color correction algorithms. The derived color correction matrices were applied to urinalysis dipstick images and compared to the spectrometer readings. Results indicated that polynomial fitting showed the lowest ∆E during calibration but failed to properly correct urine dipstick colors. Root polynomial offered the best performance in reducing color differences to be below 3 to 4 ∆E. Utilizing L*a*b values for classifying a given dipstick result according to reference concentration levels, it was found that quadratic discriminant analysis (QDA) and k Nearest Neighbor (kNN) classifiers achieved an 82.9% and 97.1% accuracy, respectively.

View contact details

UPDATE YOUR PROFILE

Is this your profile? Update it now.

Sign into your SPIE.org account

Don’t have a profile and want one?

Create an account on SPIE.org

Keywords/Phrases

Search In:

Publication Years