Abstract:
Non-negative tensor factorization can automatically find phenotypes in EHRs with minimal human supervision. In practice, the correspondence between different modalities (e.g., medications and diagnoses) is often missing, so such methods require an input tensor describing inter-modal interactions.
Heuristic methods can estimate them, but they introduce errors and produce suboptimal phenotypes. Since critical care patients have multiple diagnoses and medications, this is crucial. We propose the collective hidden interaction tensor factorization (cHITF) to jointly discover phenotypes from EHR with unobserved inter-modal correspondence.
The unobserved inter-modal correspondence is reconstructed by maximizing the likelihood of the observed matrices. Extensive experiments on the real-world MIMIC-III dataset show that cHITF infers clinically meaningful inter-modal correspondence, discovers more clinically relevant and diverse phenotypes, and outperforms several state-of-the-art computational phenotyping models in predictive performance.
Note: Please discuss with our team before submitting this abstract to the college. This Abstract or Synopsis varies based on student project requirements.
Did you like this final year project?
To download this project Code with thesis report and project training... Click Here
