Maternal Moderate Anemia Heuristics from Smartphone Image Learning Algorithms (MAHILA)

​

​

Mohammad Sarfarazul Ambiya

Chief Data Scientist

​

​

GDHD, Geneva

​

12M pregnant women in India have anemia annually

Anemia in pregnancy raises risk for maternal mortality

​

​

�

​

​

​

​

Anemia is reversible with iron supplementation

Which of these two pregnant women are anemic?

Which should receive an iron transfusion?

20 out 100 anemic patients will be missed by the WHO rapid test

(Marn, H et al Lancet 2016)

And these tests are invasive

10 out 100 anemic patients will be missed by digital hemoglobin machines (Yadav, Neogi et al)

​

​

​

And these tests are invasive

​

​

Cost $50M annually for consumables in India

Image Processing

ResNet18

Hb < 9

Hb ≥ 9

MAHILA: woman in Hindi, moderate anemia heuristics from image learning algorithms

​

​

Largest study for smartphone Hb classification for pregnant women in the world with 3000 participants

​

91-96% sensitivity

100% specificity

AUC 1

​

​

Next Steps

• Train and validate with 25,000 pregnant women

​

• Train model for general population in community

​

• Train and validate for range of smartphones

​

• Publish a foundational model for anemia classification

​

• Work with government and regulators for community-based scale-up

​

research@khushibaby.org