Artificial intelligence–driven quantification of sickle red blood cell morphology as an objective disease marker

• This newly developed artificial intelligence-based assay might serve as an objective marker of disease severity and drug efficacy. • The effects of voxelotor on erythrocyte sickling may provide an explanation for increased hemolysis after abrupt cessation of the drug. In patients with sickle cell disease (SCD), heterogeneity in erythrocyte morphology, attained via subjective interpretation with high inter-observer variability, complicates erythrocyte morphology examination for clinical monitoring and treatment evaluation. We developed an observer-independent assay that quantifies different morphologically abnormal erythrocytes grouped with high discriminatory accuracy using an automated, artificial intelligence (AI) algorithm. This assay was used to investigate whether quantification of the fractions of morphologically abnormal erythrocytes in SCD patients can be used as an objective measure of clinical severity, and to evaluate treatment efficacy. A total of 172 SCD patients were included in the analyses. At 21% O 2 , HbSS/HbSβ 0 patients appeared to have significantly higher percentages of classic sickle cells, and other aberrant cell types compared to HbSC/HbSβ + patients. In addition, a present vaso-occlusive episode (VOE) at the time of sampling was significantly associated with an increased number of aberrant erythrocyte morphologies. We also evaluated the applicability of our AI morphology-based assay for monitoring treatment effect in SCD patients, by monitoring the effect of voxelotor on sickle erythrocyte morphologies in vitro. We observed increased erythrocyte sickling at low voxelotor doses compared to both higher doses and non-treated erythrocytes. The effects of voxelotor on erythrocyte sickling observed in this study may provide a possible explanation of the recently reported clinical side effects of this drug seen in SCD patients leading to the withdrawal of the drug of the market. These results show that our AI morphology-based assay might serve as an objective marker of disease severity and treatment efficacy.