Agentic Autoresearch for CT Reconstruction

2026-07-24

AI Summary

This work addresses the gap in fair evaluation of computed tomography (CT) reconstruction methods, which often relies on idealized data that fails to reflect performance under realistic noise conditions. The authors propose the first fully automated research framework driven by a large language model (LLM) agent, capable of autonomously implementing, tuning, and benchmarking 26 reconstruction algorithms through iterative optimization—editing solvers, scheduling short-lived cluster jobs, and reading fixed metrics. Integrating a differentiable fan-beam projector, a calibrated in-FOV headroom scoring mechanism, and low-dose/sparse-view datasets, the system discovers a compact solver with only 969 parameters, achieving state-of-the-art performance on the Mayo dataset with just 0.4% of typical parameter counts. Crucially, the study reveals that rankings derived from idealized data poorly predict robustness under real-world noise: mild noise nearly inverts breast CT method rankings, though retraining substantially restores correlation (Spearman ρ rising from 0.04 to 0.61).

Abstract

Comparing CT reconstruction methods fairly is labor-intensive and largely manual, and many benchmarks use idealized data. We ask whether a large language model (LLM) agent can do the labor of reconstruction research on its own, and whether a ranking measured on ideal data predicts behavior under realistic noise. We built an agentic loop: the agent edits a solver, runs a short cluster job, reads one frozen metric, and revises. The metric is a calibrated headroom score against the FBP baseline, inside the field of view; every method shares the same differentiable fan-beam projector. We benchmarked 26 methods on Mayo low-dose CT (noise-limited) and a 128-view sparse-view breast task from the noiseless DL-Sparse-View Challenge, with validation-selected iterations scored on a held-out test set. Every trained breast model was then re-scored on noisy inputs (I_0 = 10^5 photons) without retraining, and separately retrained on matched noise. The agent independently implemented, tuned, and benchmarked all 26 methods, and recombined them into a compact solver of 969 parameters that ties the top Mayo tier at the 1% level using 0.4% of the champion's parameters. Benchmarking gives a tier of statistically indistinguishable top methods, not one winner. Mild input noise nearly inverts the breast ranking: the noiseless champion (a supervised image denoiser, hr 0.89) collapses to 0.00, while a learned primal-dual method rises to champion (0.72 to 0.93). An ideal-data leaderboard therefore does not predict robustness. The inversion is a transfer effect, not a permanent deficit: retraining on matched noise restores much of the clean ranking (Spearman rho 0.04 to 0.61). Noise is only the easiest confounder in an open-ended set (beam hardening, scatter, anatomy, disease), so no single-factor challenge certifies generality. Benchmarks should model a broad spectrum of realistic factors at once.

Problem

• CT reconstruction
• benchmarking
• realistic noise
• robustness
• generalization

Innovation

• Agentic Autoresearch
• CT Reconstruction
• Large Language Model Agent
• Noise Robustness
• Headroom Score