JMIR Publications

ABSTRACT

Background:

Anemia, defined by a reduced number or quality of red blood cells, is a widespread health condition with significant clinical implications.

Objective:

This study aimed to examine the facial spectral characteristics of anemic patients and to devise a predictive model for anemia risk utilizing machine learning methodologies.

Methods:

The study collected 285 participants from the physical examination cohorts at Shanghai Jiading Community Health Center and Shanghai Gaohang Community Health Center, from which a subset of 78 healthy controls was randomly selected. Additionally, 128 inpatients from the hematology department of Shanghai Hospital of Traditional Chinese Medicine were assessed, leading to the inclusion of 78 individuals with anemia. Spectral measurements at eight facial sites—forehead, glabellum, nose, jaw, and both zygomatic and buccal regions—were obtained using the CS-600CG spectrophotometric colorimeter. A comprehensive statistical analysis of the spectral indices' characteristics and their interrelations was performed. Subsequently, a series of ten machine learning algorithms were leveraged to formulate anemia prediction models.

Results:

The study disclosed pronounced disparities in facial spectral attributes between anemic patients and healthy counterparts. The Random Forest classifier excelled among classification models, achieving an accuracy of 0.875 (95% CI: 0.825 to 0.925) for distinguishing between anemic and control groups. The model attributed the highest predictive value to the spectral bands at nose-400nm (t=-5.797，P<0.001), right cheek-540nm(t=-4.016, P<0.001), right cheek-580nm (t=-6.121, P<0.001), right cheek-570nm(t=-6.265, P<0.001) and nose-570nm (t=5.077，P<.001). In regression analysis, the Support Vector Regression model outperformed others with an R2 value of 0.664, indicating a strong predictive capacity.

Conclusions:

Facial spectral data holds clinical significance in anemia diagnosis, and the early warning model of anemia risk constructed based on spectral information demonstrates a high accuracy rate. Clinical Trial: The studies involving human participants were reviewed and approved by the Ethics Committee of Shanghai Municipal Hospital of Traditional Chinese Medicine affiliated to Shanghai University of Traditional Chinese Medicine (registration number 2021SHL-KY-03-01) and the IRB of Shuguang Hospital affiliated with Shanghai University of TCM (registration number 2018-626-55-03).