S1063
Clinical – Upper GI
ESTRO 2026
Health, Technical University of Munich (TUM), Munich, Germany. 2 Department for Internal Medicine II, TUM University Hospital Rechts der Isar, TUM School of Medicine and Health, Technical University of Munich (TUM), Munich, Germany. 3 Arbeitsgemeinschaft Internistische Onkologie, AiO Studien gGmbH, Berlin, Germany. 4 Institute of Diagnostic and Interventional Radiology, TUM University Hospital Rechts der Isar, TUM School of Medicine and Health, Technical University of Munich (TUM), Munich, Germany. 5 Institute of Pathology, TUM School of Medicine and Health, Technical University of Munich (TUM), Munich, Germany. 6 Institute of AI and Informatics in Medicine, TUM School of Medicine and Health, Technical University of Munich (TUM), Munich, Germany Purpose/Objective: Liver transplantation is a curative treatment option for patients with early stage hepatocellular carcinoma (HCC). However, bridging treatments prior to liver transplantation are often needed to control the tumor during waiting time for a donor organ. In a prior retrospective analysis, we showed excellent histopathological tumor control aftertransarterial chemoembolization (TACE) plus stereotactic body radiation therapy (SBRT) [1]. Therefore, the aim of the TREASURE-TX (AIO-HEP-0124) trial is to prospectively evaluate efficacy and safety of TACE alone versus TACE plus SBRT in patients with early HCC. Material/Methods: Eligible patients with HCC at an early stage (BCLC A, UNOS T2) planned for liver transplantation and sufficient liver function (Child Pugh <B8) can be included in this phase II trial. We will randomize patients 1:1 into arm A (standard arm) and arm B (experimental arm). Patients in arm A will receive two TACE treatments while patients in arm B will additionally receive SBRT after TACE (see Figure 1).
collected CT images (reflecting macroscopic features like tumor size and invasion extent) and postoperative WSI slides (reflecting microscopic features like cell morphology and necrosis) for each patient. We used LLaVA and GPT-4V models to convert CT and WSI features into text descriptions (e.g., CT: "moderate thickening of the esophageal tumor wall"; WSI: "cancer cells with prominent nucleoli and rich vasculature").Model Construction: We developed a "Cross-Attention Fusion" framework. This framework uses separate CT and WSI encoders to extract macroscopic and microscopic features, respectively. These are then combined with a general text encoder and a specific text encoder to process the text prompts. The multimodal features are ultimately fused to output the pCR prediction. Results: Superior Performance: The CT & WSI multimodal model consistently outperformed the single-modality models in all centers. For example, at the Sichuan center, the CT model had an AUC of 79.20±4.21, the WSI model had an AUC of 77.23±4.55, while the multimodal model's AUC increased to 83.26±2.34. Similar trends were observed in the external validation centers (e.g., Beijing, Tianjin), where the multimodal model demonstrated consistently higher ACC, SPE, and SEN values.Survival Prediction Value: Kaplan- Meier analysis showed that patients predicted to have pCR by the multimodal model had significantly longer RFS compared to non-pCR patients. At the Xi'an center, the multimodal CT & WSI model's p-value was 0.0213, outperforming the single-modality models (CT: p=0.0046, WSI: p=0.0132). Conclusion: The multimodal model, by fusing CT images and microscopic pathological slides, effectively integrates spatial and pathological information about the tumor. This significantly improves the accuracy of pCR prediction following nCRT and provides a more effective method for stratifying patient survival risk. References: no reference Keywords: esophageal cancer pCR predictor Digital Poster 243 TREASURE-TX: Transarterial chemoembolization and stereotactic body radiation therapy in patients with hepatocellular carcinoma before transplantation Marco M.E. Vogel 1 , Ulrike Bauer 2 , Mischo Kursar 3 , Fabian Lohöfer 4 , Carolin Mogler 5 , Bernhard Haller 6 , Roland M. Schmid 2 , Stephanie E. Combs 1 , Ursula Ehmer 2 1 Department of Radiation Oncology, TUM University Hospital Rechts der Isar, TUM School of Medicine and
TACE will be performed as conventional or drug- eluting bead TACE with doxorubicin or epirubicin. In patients with good liver function (Child Pugh A5-6) the prescribed dose for SBRT should be 37.5 - 45 Gy in 3 fractions (3 fractions per week). However, in patients with impaired liver function (Child Pugh B7) or if the tumor is located near organs of the adjacent gastrointestinal tract, prescribed radiation dose should be 40 Gy in 5 fractions (3 fractions per week). Radiation dose should be prescribed to the 65% - 80% Isodose. Results: We plan to include 44 patients (22 patients in arm A;
Made with FlippingBook - Share PDF online