Retrospective Comparison of Radiation Oncology Patient Message Responses Generated by Epic and Institutionally Managed LLMs

Meiyun Cao; Federico Mastroleo; Shuoyang Wei; Yuexing Hao; Srinivas Seetamsetty; Keldon K. Lin; Madhurya Reddy; Yuzhen Ding; Jason Holmes; Jonathan E. Moonen; Shyanne L. Dobbs; Diego Santos Toesca; Mark Waddle; Wei Liu

arXiv · 2610.34464

Retrospective Comparison of Radiation Oncology Patient Message Responses Generated by Epic and Institutionally Managed LLMs

Meiyun Cao·Federico Mastroleo·Shuoyang Wei·Yuexing Hao·Srinivas Seetamsetty·Keldon K. Lin·Madhurya Reddy·Yuzhen Ding·Jason Holmes·Jonathan E. Moonen·Shyanne L. Dobbs·Diego Santos Toesca·Mark Waddle·Wei Liu

View original source ↗Read source document ↗Export metadata ↓

Abstract

We retrospectively compared Epic-generated draft replies with institutionally managed hybrid GPT-4.1 and Gemini 2.5 Pro responses to 51 selected single-turn radiation oncology messages. Six clinicians, blinded to response source, rated overall quality on a five-point scale and perceived editing effort on a four-point scale. Mixed-effects models accounted for repeated ratings by message and clinician. Adjusted overall-quality means were 2.95 for Epic, 4.27 for hybrid GPT-4.1, and 4.39 for hybrid Gemini 2.5 Pro. Corresponding editing-effort means were 2.38, 1.46, and 1.41, with lower scores indicating less perceived effort. Both hybrid models received more favorable ratings than Epic for both outcomes (all Holm-adjusted P < 0.001), without a statistically significant difference between the hybrid models. These findings support prospective evaluation of clinician-supervised draft generation, though editing time and patient outcomes were not measured, and clinical safety was not established.

Explore related subjects

physics.med-ph

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Meiyun Cao, Federico Mastroleo, Shuoyang Wei, Yuexing Hao, Srinivas Seetamsetty, Keldon K. Lin, Madhurya Reddy, Yuzhen Ding, Jason Holmes, Jonathan E. Moonen, Shyanne L. Dobbs, Diego Santos Toesca, Mark Waddle, Wei Liu. 2026-10-25. Retrospective Comparison of Radiation Oncology Patient Message Responses Generated by Epic and Institutionally Managed LLMs. https://arxiv.org/abs/2610.34464

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Just-in-time volumetric imaging: Learning 2D-3D thoracic motion models during image-guided interventions

Despite two decades of innovation, cutting-edge bronchoscopic techniques still fail to yield clear diagnostic decisions in 30% of cases and miss malignancies in 20% of cases. Intraprocedural 3D imaging drastically improves both tool positioning and diagnostic yield. However, each 3D spin represents a static snapshot of an anatomy that is continuously evolving as the patient is ventilated throughout the procedure. Here, we propose just-in-time volumetric imaging, in which a patient-specific 2D--3D motion model is trained on pre-intervention scans before it is needed and then supplies volumetric images from single fluoroscopic projections during sampling. Given the inspiration-hold CBCT already acquired for navigation and an additional motion-inclusive CBCT, our optimization regime named VoxelMap-DualScan (VM-DS), solves a low-rank motion model by alternating least squares and then amortizes it into a projection encoder trained on synthesized projections with known motion, all within a 10-minute window. On eight held-out cases from the public DIR-Lab 4DCT dataset, VM-DS achieved a mean landmark target registration error of 2.8 mm, a structural similarity index of 0.95 and a peak signal-to-noise ratio of 37.6 dB. Accuracy was maintained when reducing the number of projections used for training from 227 to 57. Inference takes 1.5 ms per projection, making the framework suitable for real-time use during image-guided interventions. Code will be made available at https://github.com/Image-X-Institute/Voxelmap/tree/VM-DS

physics.med-ph↗

arXiv2026

Gold nanorods targeting tumor vasculature: Monte Carlo evaluation of the radioenhancement effects

Gold nanoparticles targeted to tumor vasculature can raise the dose to endothelial cells, and the enhancement depends on their position relative to the vessel wall. Despite growing interest in endothelial radiosensitization, intravascular and immediate perivascular dose enhancement has not been resolved at the level of individual endothelial cells for different nanoparticle locations. Earlier vascular dosimetry scored thin wall layers, analytic endothelial slabs, or tissue-region averages. This study quantifies the dose enhancement ratio (DER) of gold nanorods (AuNRs) in individual endothelial cells and in the perivascular region for three AuNR locations. Track-structure Monte Carlo simulations in TOPAS modeled a 10 um long, 16 um diameter microvessel segmented into six endothelial cells. AuNRs were placed homogeneously in the lumen, on the inner vessel wall (vascular-targeted), or on the outer vessel wall (extravasated) at concentrations of 5 to 50 mg Au/g water within the vessel volume under 50 and 100 kVp X-rays. All DERs were corrected for the lack of charged-particle equilibrium (CPE) inherent to microscale simulations. Inner wall-attached AuNRs produced the highest microvessel-averaged DERs, reaching approximately 3.8 at 50 mg Au/g water and approximately 1.96 at 18 mg Au/g water under 50 kVp. The DER increased linearly with AuNR concentration and was higher at 50 kVp than at 100 kVp. Including AuNRs outside the simulated segment (infinite vessel approximation) increased endothelial DERs by 5 to 10 percent. In the perivascular region, the DER was highest within the first 10 um, where outer wall-attached AuNRs gave the highest values. The results establish a physical dose enhancement basis for endothelial and perivascular cells and can guide the design of vascular-targeted radiosensitization.

physics.med-ph↗

arXiv2026

At last, an explicit singular value decomposition for electrical impedance tomography

We derive a diagonalization of the normal operator of the linearized electrical impedance tomography (EIT) problem on the disk in the case of a radially symmetric Schrödinger potential with Dirichlet-to-Neumann data. The diagonalization is achieved by transforming the normal operator to the adjacent interval finite Hilbert transform which has a known spectral theory including a commuting differential operator. The normal operator in this case is not compact and the spectrum continuous.

physics.med-ph↗