ESTRO 2025 - Abstract Book

S119

Invited Speaker

ESTRO 2025

loco-regional RT, and no significant differences in oncological outcome data were found between the arms at 5 years after RT. Moreover, the 5-year loco-regional recurrence rate was very low, 1.6%. Although these studies yielded interesting results, an ALND is nowadays only rarely performed, such that studies where only SARP has been done are gaining interest. Ad 2) In case of SARP only, it is important to discriminate the several combinations of cN and ypN status, i.e. cN0/ypN0(SARP), or cN+/ ypN0 (SARP), or cN0/ ypN+ (SARP), or cN+,/ypN+ (SARP). For these different scenarios, a variety of approaches is and has been investigated, varying from no further treatment of the axilla at all, to replacing an ALND by less or more extensive regional RT. During the presentation a review will be given on the available data up till now, mostly obtained from retro or prospective cohort studies, and of ongoing trials. References: 1 Buchholz et al. Pathologic tumor size and lymph node status predict for different rates of locoregional recurrence after mastectomy for breast cancer patients treated with neoadjuvant versus adjuvant chemotherapy. Int J Radiat Oncol Biol Phys. 2002 Jul 15;53(4):880-8. doi: 10.1016/s0360-3016(02)02850-x. 2 EBCTCG (Early Breast Cancer Trialists' Collaborative Group). Effect of radiotherapy after mastectomy and axillary surgery on 10-year recurrence and 20-year breast cancer mortality: meta-analysis of individual patient data for 8135 women in 22 randomised trials. Lancet. 2014 Jun 21;383(9935):2127-35. doi: 10.1016/S0140-6736(14)60488-8. Epub 2014 Mar 19. Erratum in: Lancet. 2014 Nov 22;384(9957):1848. 3 de Wild et al. De-escalation of axillary treatment in the event of a positive sentinel lymph node biopsy in cT1-2 N0 breast cancer treated with mastectomy: nationwide registry study (BOOG 2013-07). Br J Surg. 2024 Apr 3;111(4):znae077. doi: 10.1093/bjs/znae077. 4 Manoumas et al. Locoregional irradiation in patients with biopsy proven axillary node involvement at presentation who become pathologically node negative after neoadjuvant chemotherapy: primary outcomes of NRG oncology/NSABP B-51/RTOG 1304. SABCS 2023.

4846

Speaker Abstracts Foundation models in radiotherapy Harini Veeraraghavan Medical Physics, Memorial Sloan Kettering Cancer Center, New York, USA

Abstract:

Increasing availability of large numbers of image datasets combined with the unprecedented computing capabilities have led to the development of large image foundation models. Recent works have shown capabilities to create medical image foundation models often using publicly available medical image datasets specific to individual modalities or disease sites, as well as for specific types of tasks such as segmentation. In general, foundation models are created from large numbers of diverse datasets in order to extract useful feature representations that can then be easily applied to a variety of downstream tasks with little to no finetuning. As a result, these models are highly attractive for radiotherapy, which involves several steps that benefit from automation and requires analysis of variety of imaging modalities. In this talk, I will briefly present some common foundation models and pretraining approaches, methods that extend single to multi-modal foundation models, and applications in various radiotherapy tasks including segmentation and classification. I will also present some considerations for pretraining such models when training for downstream tasks involving segmentation versus classification, show how pretraining impacts feature reuse and downstream labeled data requirements for fine-tuning and refinement.