Evaluating Deep Human-in-the-Loop Optimization for Retinal Implants Using Sighted Participants

2025-01-31

AI Summary

Existing computational models of artificial vision exhibit discrepancies with human perceptual decision-making, limiting the clinical translation of retinal prostheses. Method: This study presents the first empirical validation of deep human-in-the-loop optimization (HILO) for personalizing stimulation strategies in retinal prostheses using real human participants. We employed sighted subjects simulating artificial vision, integrated a deep stimulation encoder (DSE), and applied a paired-comparison preference paradigm within a closed-loop iterative optimization framework. Performance was evaluated under three robustness conditions: standard settings, mis-specified detection thresholds, and out-of-distribution sampling. Results: HILO-optimized stimulation policies significantly outperformed both baseline encoders and unoptimized DSE across all conditions (log-odds significantly positive), demonstrating clinical adaptability. Crucially, this work establishes the first human-feedback-driven validation of HILO, reveals critical disparities between simulated and actual perceptual decisions, underscores the necessity of preclinical human evaluation, and provides a generalizable methodological framework for personalized visual prosthesis fitting.

Technology Category

• Humans and AI: Human-in-the-loop Machine Learning
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Fairness and ethical considerations in crowd work and in human-in-the-loop AI systems

Abstract

Human-in-the-loop optimization (HILO) is a promising approach for personalizing visual prostheses by iteratively refining stimulus parameters based on user feedback. Previous work demonstrated HILO's efficacy in simulation, but its performance with human participants remains untested. Here we evaluate HILO using sighted participants viewing simulated prosthetic vision to assess its ability to optimize stimulation strategies under realistic conditions. Participants selected between phosphenes generated by competing encoders to iteratively refine a deep stimulus encoder (DSE). We tested HILO in three conditions: standard optimization, threshold misspecifications, and out-of-distribution parameter sampling. Participants consistently preferred HILO-generated stimuli over both a na""ive encoder and the DSE alone, with log odds favoring HILO across all conditions. We also observed key differences between human and simulated decision-making, highlighting the importance of validating optimization strategies with human participants. These findings support HILO as a viable approach for adapting visual prostheses to individuals.

Problem

• Deep Human-in-the-Loop Optimization
• Visual Prosthesis
• Personalized Stimulation Strategies

Innovation

• Human-in-the-Loop Optimization
• Visual Prosthesis Personalization
• Stimulus Parameter Adjustment