Fig. 9: Overview of the WEST phenotyping pipeline. | npj Digital Medicine

Fig. 9: Overview of the WEST phenotyping pipeline.

From: A weakly supervised transformer for rare disease diagnosis and subphenotyping from EHRs with pulmonary case studies

Fig. 9: Overview of the WEST phenotyping pipeline.

The schematic illustrates the end-to-end workflow of WEST. (1) Cohort identification and labeling assign gold-standard (expert-validated) and silver-standard (probabilistic) labels. (2) EHR sequence pre-processing converts longitudinal structured and unstructured data into aggregated concept sequences with frequency encoding. (3) A transformer encoder models dependencies among clinical concepts. (4) Feature pooling and fine-tuning generate patient-level phenotype predictions and low-dimensional embeddings for subphenotyping. Figure created using Canva.

Back to article page