# Automated Image Analysis for Prevention of Radiotherapy Delivery Errors

> **NIH AHRQ R01** · UNIVERSITY OF CALIFORNIA LOS ANGELES · 2020 · $322,450

## Abstract

Project Summary/Abstract
Approximately 1 million patients per year receive radiation therapy in the United States as part of their cancer
care. Rapidly advancing technology has over the past two decades enabled treatments in which irradiated
volumes are made to conform ever more tightly to tumors. Increasing conformality has resulted in better cure
rates and lower side effects, but requires highly precise patient positioning with little margin for error. Patient
positioning is performed by radiation therapy technologists, usually by visually aligning on-board x-ray setup
images to the CT scans used to plan patients' treatments. Academic studies, public records, and works of
investigative journalism have demonstrated that, despite quantitative positioning processes and extensive and
rigorous quality control, human error leads to so-called never events: treatments with serious alignment errors
or with the wrong patient's treatment plan. Based on never events reported at UCLA, at least 1,400 such
events occur nationally per year.
Though rare, radiotherapy never events have potentially devastating consequences, and reducing their
occurrence is strongly motivated. Radiotherapy is already subject to intensive quality control procedures, so
further reduction is likely best achieved through automation in order to avoid additional burden on an already
labor-intensive workflow. This project will develop an automated, on-line never event prevention system
(NEPS) that will interlock the radiotherapy machine to prevent treatment if the patient is not correctly aligned or
if the wrong patient plan is loaded, reducing never events by an order of magnitude and directly addressing the
AHRQ priority of improving patient safety. Additionally, this project will retrospectively measure the never event
rate at UCLA and Veteran's Health Administration (VHA) radiotherapy clinics, testing the hypothesis that
radiotherapy never events are significantly under-reported.
In Aim 1, a planar x-ray-based never event detection algorithm will be developed, expanding on an existing
volumetric-image based never event detection algorithm. It will be shown that the planar x-ray algorithm has a
sensitivity of at least 90% and a specificity of 99%. In Aim 2, the automated never event detection algorithms
will be retrospectively deployed to measure the actual never event rate, which is hypothesized to be
significantly greater than the reported never event rate, by analyzing over 250,000 setup images from existing
UCLA and VHA clinical image databases. In Aim 3, the NEPS will be developed and deployed in a UCLA
radiotherapy suite and assessed in terms of number of never events detected, false positive rate, costs of
system implementation including personnel costs to address false positives, and potential unintended
consequences. This Aim will provide important information regarding feasibility for broader dissemination and
implementation. If successful, this project will ultimately lead ...

## Key facts

- **NIH application ID:** 9984393
- **Project number:** 5R01HS026486-03
- **Recipient organization:** UNIVERSITY OF CALIFORNIA LOS ANGELES
- **Principal Investigator:** James Michael Lamb
- **Activity code:** R01 (R01, R21, SBIR, etc.)
- **Funding institute:** AHRQ
- **Fiscal year:** 2020
- **Award amount:** $322,450
- **Award type:** 5
- **Project period:** 2018-09-30 → 2023-07-31

## Primary source

NIH RePORTER: https://reporter.nih.gov/project-details/9984393

## Citation

> US National Institutes of Health, RePORTER application 9984393, Automated Image Analysis for Prevention of Radiotherapy Delivery Errors (5R01HS026486-03). Retrieved via AI Analytics 2026-05-24 from https://api.ai-analytics.org/grant/nih/9984393. Licensed CC0.

---

*[NIH grants dataset](/datasets/nih-grants) · CC0 1.0*