Investigation of predictability of photo images from different anatomic regions for detecting anemia among preterm infants Open Access
Du, Chenxi (Spring 2023)
Abstract
Background: Anemia is a global health issue affecting people of all ages, with the highest prevalence among preschool-age children. Traditional diagnosis methods for neonatal anemia involve frequent blood sampling, which can lead to blood loss-induced anemia and the development of complications. Therefore, a non-invasive method for diagnosing anemia for preterm infants is needed. One potential solution is developing a smartphone application that captures videos and images through the camera and predicts Hgb levels based on RGB values. Several smartphone applications have been developed for monitoring hemoglobin levels in adults, but there is no evidence regarding their efficacy in infants. Objective: The purpose of this study is to develop a method for predicting neonatal anemia among preterm infants using RGB values and metadata extracted from smartphone images of various body regions, including the fingernail, toenail, and palm. Furthermore, we aim to determine which body region that can predict anemia with better degree of accuracy. Method: PCA was conducted on RGB data extracted from body region images for dimension reduction. Four logistic regression models were built to examine for the best region for predicting anemia. Stepwise model selection was employed to select the proper predictors among image metadata (PCs of RGB value, brightness value, exposure time) and infants’ demographic data (age, gender, race, ethnicity, birth weight, and gestational age). Cross-validation was used to test accuracy and AUC is the main criterion. Result: 65 infants and 1000 images from 6 body regions were included in the analysis. Principal component analysis was used to include image data in the models, and four principal components were selected. Logistic regressions were fitted separately for the whole dataset and regional datasets. The Palm model reported the highest AUC comparing with all other models with cross validation AUC as 0.726, while the Fingernail Model has the lowest AUC (0.663). Conclusion: The palm region images gave a slightly better result comparing with other anatomic regions. However, in practice, there are difficulties for nurses taking pictures in this vulnerable population. Given that there is no large difference in the AUC, in practice, any region can be chosen.
Table of Contents
1. Introduction. 1
2. Method. 4
2.1 Data Source. 4
2.1.1 Cohort Selection. 4
2.1.2 Image Data Collection. 4
2.2 Statistical Analysis. 6
2.2.1 Principal Component Analysis. 6
2.2.2 Logistic Regression Model 7
3. Result 9
3.1 Preliminary Analysis. 9
3.2 Prediction results. 10
4. Discussion. 13
5. Tables: 15
6. Appendix: 24
7. Reference: 26
About this Master's Thesis
School | |
---|---|
Department | |
Subfield / Discipline | |
Degree | |
Submission | |
Language |
|
Research Field | |
Keyword | |
Committee Chair / Thesis Advisor |
Primary PDF
Thumbnail | Title | Date Uploaded | Actions |
---|---|---|---|
Investigation of predictability of photo images from different anatomic regions for detecting anemia among preterm infants () | 2023-04-10 09:55:24 -0400 |
|
Supplemental Files
Thumbnail | Title | Date Uploaded | Actions |
---|