DeepProbLog for Stroke Detection from Incomplete Medical Data
A new arXiv paper (2608.08561) explores the use of DeepProbLog, a neuro-symbolic framework, for diagnostic reasoning in stroke detection from multimodal data. The research addresses the challenge of incomplete information in medical applications, where privacy concerns limit raw data availability, making summary statistics from literature crucial. The proposed workflow integrates deep learning components for analyzing patient images within a transparent probabilistic logic programming framework under the distribution semantics. It employs maximum entropy techniques to complete probabilistic information and uses ProbLog 2 for probabilistic reasoning. The study demonstrates a pathway from literature-derived summary statistics to a functional diagnostic system, highlighting the potential of neuro-symbolic AI in medical diagnostics.
Key facts
- Paper arXiv:2608.08561, announced as new.
- Focuses on stroke detection from multimodal data.
- Uses DeepProbLog, an extensible neuro-symbolic approach.
- Addresses privacy concerns by using summary statistics from literature.
- Employs maximum entropy techniques to complete probabilistic information.
- Utilizes ProbLog 2 for probabilistic logic programming.
- Case study demonstrates diagnostic reasoning from incomplete information.
- Combines connectionist components for image analysis with probabilistic reasoning.
Entities
Institutions
- arXiv