How it works

We have trained several models that would suggest the most probable diagnoses based on age, gender and blood laboratory results. To get these models to:

• a) cover enough diagnoses to be useful and
• b) be precise enough in the predictions

the number of required parameters exceeds 60 for a model covering only 65 hematological relevant diagnoses. While we still offer access to these models for interested parties this is not not useful for everyday clinical practice, which is why we split our model into 2 main steps.

Based on a set of only age, gender, symptoms and 20 basic blood laboratory results the Model A determines the most probable disase category for a patient. This category contains a collection of diagnoses and a further evaluation requires more laboratory measurements that are also suggested by the model.

Taking these additional parameters plus the initial input Model B that was specifically trained on only diagnoses in the given disease category suggests the most likeley diagnoses for a patient.

Currently a lightweight version of Model A is available for testing (Cluster Model). This model is constantly improved and extended until summer 2024. Similarly we are currently working on several versions of Model B (one model for each diagnosis category) and on the definitions of the additionally needed parameters. One of the these models with a focus but not a specialization on hematology (30B Model) is available for testing. We do not expect to finish all models before autum 2024 due to ongoing data cleaning efforts and forseen additional data.

Initial Input

• Age, Gender,
• Symptoms and
• 31 Blood Lab Results

Model A

Based on the initial input this model suggests the most likely disease category and additional laboratory measurements to further narrow the range of possible diagnoses.

Most Probable Diagnosis Category

• Coagulation
• Kidneys
• Abdomen/Liver
• Hematology
• Heart/Muscle
• Infection
• Metabolism

Required Parameters for Model B

• microscopic blood count,
• RET-He,
• erythropoietin,
• folic acid,
• soluble transferrin receptor

Model B

Based on the initial input and the additional parameters this model suggests the most likely diagnoses for a patient.

Most Likely Diagnoses