KidneyCare Clinical Assistant
An explainable clinical decision-support system for nephrology, built with Kyung Hee University Hospital, classifying patients into 33 physician-validated clinical situations.
.png&w=1920&q=75)
Comprehensive Technical Write-Up Available
Explore the peer-reviewed methodology, system architecture diagrams, benchmark comparisons, and academic citations.
Project Overview
KidneyCare Clinical Assistant is a research-driven clinical decision support system developed as an SEECS, NUST industry project in collaboration with Kyung Hee University Hospital, supervised by Dr. Syed Imran Ali. It engineers an explainable, rule-based decision-support engine for nephrology that translates physician-validated clinical logic into deterministic classification across 33 clinical situations derived from clinical groups, PTH thresholds and trends, corrected calcium, and phosphate levels. The system handles real clinical constraints such as missing, outdated, or conflicting lab data, and performs longitudinal comparisons against each patient's most recent prior visit, enforcing per-test validity windows so recommendations stay grounded in genuinely comparable data.
Problem It Solves
Nephrologists must interpret multiple lab values across time while managing follow-up visits and treatment decisions. Manual analysis of reports, especially when lab data is missing or outdated, increases cognitive load and risk of oversight. There was a need for a system that centralizes patient history, validates lab relevance, and supports consistent, guideline-based decision-making that is explainable enough for physicians to trust and safely override.
Tools & Technologies Used
Skills Involved
Challenges
Designing medical logic that is safe, explainable, and aligned with real hospital workflows was the biggest challenge. The system had to correctly handle edge cases such as missing tests, expired lab validity, and partial reports, while still providing useful recommendations. Translating clinical reasoning into deterministic software logic covering 33 distinct clinical situations required close iteration with nephrology instructors and careful validation against clinical groups, PTH thresholds and trends, corrected calcium, and phosphate levels.
Learnings
This project provided hands-on experience in building safety-critical, explainable software. I learned how clinical decision support differs from typical applications, the importance of explainability and physician trust in medical systems, and how to convert domain expertise into structured, longitudinally-aware logic. It also strengthened my ability to collaborate with non-technical domain experts and design systems meant for real clinical use rather than demos.