Precision Radiation Equity: A Closed-Loop AI Framework for Autonomous CT Dose Optimization In Resource-Constrained Clinical Environments
Abstract
Purpose
Standard dose optimization fails in low-resource settings due to unique patient phenotypes, heterogeneous aging equipment, and absent validation pathways. This work presents a closed-loop digital twin ecosystem that autonomously synthesizes institution-specific ultra-low-dose CT images and iteratively refines clinical protocols, achieving precision dose reduction without patient exposure or vendor dependency.
Methods
The Autonomous Protocol Optimization Network (APOLLO), a physics-embedded generative AI trained on 2,317 routine clinical scans from Nigerian center. APOLLO uniquely integrates four conditional inputs: local scanner performance degradation curves , regional patient body composition distributions, specific detector quantum efficiency models, and local grid power stability patterns. Validation employed a three-phase prospective implementation: Task-transfer-function and 3D noise-power-spectrum analysis comparing APOLLO-synthesized 0.8 mSv images against actual 0.8 mSv acquisitions on three scanner generations (correlation >0.99, p<0.001). A randomized, blinded multi-reader multi-case study evaluating diagnostic accuracy for subtle hepatic and pancreatic pathology using receiver-operating-characteristic analysis..Implementation of APOLLO-recommended protocols across four Nigerian hospitals with continuous dose monitoring over eight months.
Results
APOLLO achieved diagnostic non-inferiority at 0.8 mSv compared to standard 4.0 mSv protocols (ΔAUC = 0.007, 95% CI: -0.018, 0.032). The system identified and corrected a 32% overdose in our pediatric abdominal protocol that vendor software had missed. Post-implementation, the median institutional CTDIvol decreased by 41% (15.2 to 9.0 mGy) while repeat-scan rates remained unchanged (3.1% vs 3.0%, p=0.87). Inter-institutional adoption demonstrated consistent dose reductions of 28-44% across all sites.
Conclusion
This work establishes an autonomous, self-optimizing radiation safety framework specifically engineered for low-resource clinical reality. APOLLO transcends generic dose reduction by creating a continuous improvement loop that adapts to local infrastructure and population characteristics. The system provides immediate, measurable population-level risk reduction while creating a scalable model for global radiation equity, representing a paradigm shift from passive protocol adoption to active, intelligent safety optimization where it is most critically needed.