The present application relates to the field of medical
artificial intelligence and clinical auxiliary decision-making technology, in particular to an
abnormal test result combination
pattern recognition method and
system based on
deep learning, comprising: receiving test results output by a hospital test
information system to construct a 48-dimensional test index vector; calculating a multi-dimensional joint deviation degree based on a group health joint distribution
reference model; constructing a patient individual baseline with stable period test values for
baseline drift correction, and inferring an individual baseline and giving a baseline disturbance double-channel representation when the stable period is missing with a meta-learning baseline
inference subnetwork output; inputting the individualized joint deviation degree and the baseline disturbance double-channel representation into a
multilayer perceptron network classifier to output five types of severe clinical event prediction probabilities of
sepsis,
acute kidney injury, disseminated intravascular coagulation, acute
liver failure and acute
exacerbation of chronic diseases, and the training loss contains a pathogenic causal diagram prior constraint term; when the
prediction probability exceeds 40%, an orange reminder is pushed to the mobile terminal of the responsible nurse.