ESTRO 2025 - Abstract Book

S3

Invited Speaker

ESTRO 2025

virtual RTstructs, respectively) based on predefined input parameters. The applications were developed within a DevOps framework, utilizing Microsoft Azure, with dedicated teams of two to three full-time equivalents (FTEs).

All applications were built around a microservice infrastructure, aimed at scalability, maintainability, and the reusability of software. The total development time was 2 years for AI autocontouring for brachytherapy, and 0.7 years for SmartAdapt. These are the estimated development costs for the two applications, with MDR compliance included in the development estimates: AI autocontouring brachy SmartAdapt

Personnel costs

€200,000 (1 FTE, 2 y)

210,000 (3 FTE, 0.7 y)

Cloud computing and storage

€40,000

€21,000

Total development cost

€240,000

€231,000

Annual maintenance

€25,000

€25,000

A direct comparison of these solutions to commercially available ones is difficult. However, looking at applications in RT with similar complexity, e.g. auto-segmentation, or automatic DICOM routing modules, the cost for a perpetual site license would be on the order of €50,000 - €150,000, and annual service costs are typically on the order of 5 15% of that. Based on these estimates, in-house development is likely no longer the most cost-effective option when considering financial investment alone. Furthermore, the added benefit of using commercial solutions is the guarantee of a support organization backing the application, whereas maintaining full support for in-house developed solutions is challenging. No. Despite financial drawbacks, there are strong arguments in favor of continuing in-house development, especially when it enables pushing beyond the state of the art. Many groundbreaking innovations in radiation therapy (RT) originated from in-house efforts before becoming industry standards. For example, inverse planning, portal imaging, CBCT based image guidance, VMAT, portal dosimetry, and the MR-linac, all started as research initiatives in clinics before becoming mainstream technology. These advancements may never have emerged if institutions had relied solely on commercial vendors, whose interests are mainly market-driven. Furthermore, while the cost comparison with commercial software is important, if in-house development enables a large efficiency step which is not available through commercial means, its development may still be cost effective. For example, our AI autocontouring for brachy development, which is not commercially available yet, reduces the contouring time by 30 minutes per patient. However, to maximize impact and long-term feasibility, in-house development should ideally be pursued in collaboration with manufacturers, ensuring that pioneering technologies can transition into widely available products. Furthermore, such collaborations may help in financing the initial investment of the in-house development. Should we abandon in-house development?

Conclusion

Ultimately, in-house software development is most likely not cost-effective under current regulatory constraints. However, it remains strategically important when it facilitates innovation or efficiency beyond existing commercial solutions.