Poster Poster Program Therapy Physics

Stress Testing Cross Domain Generalization of High Field Brain Metastasis Auto Contouring on 0.55T MRI

Abstract
Purpose

To quantify volumetric reliability limits of a vendor brain metastasis auto contouring prototype trained only on 1.5T and 3T MRI when deployed on 0.55T MRI.

Methods

Twelve patients with 95 physician contoured metastases were imaged on a 0.55T MR simulator. Lesion volumes ranged from 0.014 to 18.9 cc. The locked commercial prototype was applied in a zero-shot setting with no exposure to less than 1.5T data. Lesion level detection was defined as any overlap between matched physician and AI contours. Geometric fidelity was summarized using Dice similarity coefficient. Performance was stratified by volume into a gray range from 0.014 to 0.065 cc and a reliable range above 0.065 cc. False positives were summarized across the same thresholds.

Results

Overall sensitivity was 92.6% (88/95). Sensitivity in the reliable range was 94.4% (67/71) with mean Dice 0.83. In the gray range, sensitivity was 87.5% (21/24) with mean Dice 0.58. For small lesions, sensitivity was 87.9% (29/33) for volumes below 0.1 cc and 91.9% (34/37) for 0.1 to 0.5 cc. Seven lesions were missed with mean volume 0.10 cc, including four above 0.065 cc. The model produced 57 false positives, 4.8 per patient, with 86.0% (49/57) below 0.065 cc and 38.6% (22/57) below 0.014 cc.

Conclusion

High field trained auto contouring generalized to 0.55T MRI with high sensitivity for lesions at or above 0.1 cc and stable performance above 0.065 cc. The dominant limitation was increased low volume false positives and reduced Dice for small lesions. The analysis of false positives is complicated by the presence of previously treated lesions in 8 patients of which several were enhancing in post-contrast imaging. However, a post processing volume filter near 0.065 cc can substantially reduce false positives, with the operational threshold guided by institutional tolerance for missing very small disease.

People

Related

Similar sessions

Poster Poster Program
Jul 19 · 07:00
Python-Based Automation Framework for Annual Machine QA Data Archiving In Qatrack+

Annual water-tank measurements help ensure beam characteristics remain consistent with commissioning baselines. However, the lack of a standardized processing workflow and decentralized data storage makes it difficult to analyze...

Syed Bilal Ahmad, PhD
Therapy Physics 0 people interested
Poster Poster Program
Jul 19 · 07:00
User Expectations and Current Availability of HDR Brachytherapy Audits In Europe

The aim of this work was to evaluate the need to implement more dosimetric audits in high‐dose‐rate brachytherapy (HDR-BT) in Europe and to identify which characteristics such audits should meet according to users.

Javier Vijande, PhD Laura Oliver Cañamás
Therapy Physics 0 people interested