Clinical decision support system for the interpretation of blood lab results and diagnosis based on them

Interpreting laboratory results and deriving a diagnosis from them is a common and demanding task in medicine. The team has developed a clinical decision support system (CDSS) that assists doctors in this task and in selecting relevant measurements for a suspected diagnosis.
This CDSS is based on machine learning models that have been trained with medical data from the University Hospital of Zurich over the last 12 years. Based on age, gender, symptoms and blood laboratory results, the software determines the most likely diagnoses for a patient.

The process is adapted to clinical practice and takes place in two steps:
In the first step, a model takes 31 basic blood parameters (including relative and absolute values of some measurements) and symptoms to suggest a diagnostic group and further relevant measurements to narrow down the possible diagnoses.
In the second step, and with the results of further laboratory measurements, the CDSS applies specialised models to suggest the patient’s specific diagnoses.

The first-stage model achieved an accuracy (F1 score) of 90%, while the specialised models range from 71% to 94%.
In the course of the project, the model was expanded to include automatic blood cell classification. This enables better differentiation between reactive and neoplastic diseases.