Deep learning helps assess stored blood quality

Researchers report that a neural network trained by experts provides 76.7% agreement with experts in identifying the morphology of damaged red blood cells in stored blood; the authors also report that allowing the network to learn to detect damage independently of experts and training using only the storage duration of each blood sample yielded better predictions of blood quality compared with prediction by experts.

Article #20-01227: "Objective assessment of stored blood quality by deep learning," by Minh Doan et al.

MEDIA CONTACT: Anne E. Carpenter, Broad Institute, Cambridge, MA; tel: 617-714-7750; e-mail: anne@broadinstitute.org; Michael Kolios, Ryerson University, Toronto, CANADA; e-mail: mkolios@ryerson.ca

###