Citations

Abstract Background An accurate discrete spot scanning (DSS) beam delivery time (BDT) model is essential, particularly as 4D dynamic dose accumulation calculations rely on machine‐specific time models for reliable interplay evaluation. Purpose To derive a BDT model for a synchrotron‐based Hitachi Probeat PBS proton system from oscilloscope measurements of key BDT structures, including multi‐energy extraction (MEE) characteristics. Methods Key BDT structures were characterized using oscilloscope measurements, supplemented by delivery log files where appropriate. Spot‐to‐spot scanning trajectories were visualized using a CROSSmini 2D strip ionization chamber (Liverage Biomedical Inc, Taiwan) via the dose‐driven continuous scanning (DDCS) mode. The MEE design and MEE charge recapture efficiencies were also characterized. In addition, the impact of the number of delivered spots on both the extractable fraction of charge and MEE recapture efficiency was investigated. Results An improved synchrotron‐based BDT model was derived that incorporates dead times between radiofrequency knockout (RFK) and high‐speed steering magnet (HSST) signals. Spot‐to‐spot movement follows a “hockey‐stick”‐like behavior where the effective spot‐to‐spot move time is governed by the slower axis. An energy‐dependent variation in the number of MEE layers deliverable per spill was also observed, with fewer MEE layers per spill (1–3) was observed in low and high energy ranges compared to the mid‐energy range (4–5), deviating from the nominal vendor value of five MEE layers per spill per block. A reduction in the extractable fraction of charge was observed with an increasing number of spots per spill, while increasing the number of delivered spots in the preceding layer significantly reduced the MEE recapture efficiency. Conclusion MEE characteristics should be incorporated into synchrotron‐based time models for accurate beam delivery modeling. The methodology presented here also provides a framework for centers seeking to derive machine‐specific delivery time models for their proton systems.

Objective Data on cardiac index (CI) in patients undergoing coronary angiography (CAG) can enhance decision-making and improve clinical outcomes. However, this necessitates an additional invasive procedure, and therefore, it is frequently omitted. This study explores the novel application of convolutional neural networks (CNNs) to predict CI from routine CAGs. Methods and analysis Patients who underwent CAG and simultaneous right heart catheterisation at Mayo Clinic between 2002 and 2023 were included. A three-dimensional (3D)-CNN model (X3D architecture) was developed to predict binary CI categories (normal: CI≥2.2 vs abnormal: CI<2.2 L/min/m²) from CAG videos. CAGs were randomly split into training (70%), validation (15%) and test (15%) sets. All available videos from each study were entered into the model. The model’s performance was evaluated using the area under the curve (AUC), sensitivity, specificity and F1 score. Results A total of 15 297 CAG studies were considered. After applying a quality-check algorithm to exclude low-quality or irrelevant CAGs, 11 475 studies from 9103 patients were included. CI was abnormal in 25.2% of studies. The model achieved an AUC of 0.83 (95% CI 0.81 to 0.85), sensitivity of 0.72 (95% CI 0.68 to 0.77), specificity of 0.78 (95% CI 0.75 to 0.80) and F1 score of 0.62 (95% CI 0.58 to 0.65) on the test dataset. Limiting the input to 4, 3 or 2 prespecified CAG videos representing standard views did not significantly affect the model’s performance, supporting its applicability in varied clinical workflows. Conclusions This study demonstrates the feasibility of leveraging deep learning models to predict CI from CAG videos. The proposed 3D-CNN model shows promise as a tool for providing physicians with real-time data to aid decision-making during CAG procedures.

Abstract Early-phase clinical trials of BTK degraders in relapsed/refractory CLL have shown response rates of over 80%, including in patients with BTK mutations that confer resistance to conventional BTK inhibitors (BTKi). The molecular basis of resistance to BTK degraders is unknown and, given the promise of this class of drugs, would hold implications for therapy sequencing and the design of next-generation BTK degraders. To this end we identified five patients at Memorial Sloan Kettering who developed progressive disease while being treated on the phase 1 trials of zelebrudomide (NCT04830137, 4 patients) or bexobrutideg (NCT05131022, 1 patient). In two patients, one treated with zelebrudomide and the other with bexobrutideg, a pre-existing BTK A428D mutation had expanded at progression, rising from &lt;1.5% at the start of treatment to &gt;25%. In both cases, the increase in frequency of BTK A428D coincided with the loss of BTK degradation in peripheral blood B cells. No mutation in cereblon, the ligase utilized by these BTK degraders, was detected. To validate these clinical findings we generated wild-type and BTK A428D knock-in single-cell clones in TMD8 and HBL1 cells. BTK A428D conferred resistance to BTKi, including ibrutinib, acalabrutinib, zanubrutinib, and pirtobrutinib, as well as the clinical-stage BTK degraders zelebrudomide, bexobrutideg, BGB-16673, and AbbVie-1 (Compound 1, WO 2023183811). BTK A428D’s pan-resistance is unique compared to previously studied BTKi resistance mutations which remain susceptible to at least one BTKi and all clinical-stage BTK degraders tested. To understand how BTK A428D confers pan-resistance to BTKi and BTK degraders we conducted surface plasmon resonance studies which demonstrated that BTKi and all four clinical-stage BTK degraders tested are unable to bind BTK A428D. We next solved the crystal structure of apo BTK A428D to 1.8Å, which revealed that the mutant aspartate, which is larger and negatively charged compared to the wild-type alanine, sterically blocks BTKi and degraders from accessing residues in the hinge region of the kinase that are required for drug binding. We next asked why BTK A428D, despite conferring pan-resistance, has only been detected in 1.1% of patients treated with BTK inhibitors. Intriguingly, we noted that in one of our index patients, the frequency of BTK A428D declined and became undetectable after the withdrawal of zelebrudomide, suggesting a fitness defect. To assess this hypothesis, we compared the proliferation rates of five TMD8 wild-type and five BTK A428D single-cell clones; the BTK A428D clones grew more slowly and were consistently outcompeted in mixing studies. In summary, low-frequency BTK A428D mutations can expand under the selection pressure of BTK degraders and confer clinical resistance by preventing drug binding. The mutation, however, is outcompeted in the absence of BTK-targeted therapies. These findings provide rationale for the investigation of combination therapies and the development of next-generation BTK degraders capable of circumventing BTK A428D. Citation Format: Quinlan Sievers, Hao Lu, Allison Cool, Stefan Grajewski, Tim Kong, Mark Noviski, Xiaoli Mi, Emel Ficici, Ratul Mukerji, James Iuliano, Jordan Ye, Mateo Sanchez Garcia de los Rios, Hugo Bousquet, Sarah Whelan, May Tan, Christopher Phelps, Jahan Raman, Ryan Notti, Lizbeth Mendoza Navarrete, Samantha Sekeres, Eria Lamkin, Eduardo Bravo, Mark Ewalt, Prioty Islam, Anthony Mato, Lindsey Roeker, Justin Taylor, Meghan Thompson, Gwenn Hansen, Omar Abdel-Wahab. BTK A428D is a recurrent pan-resistance mutation to BTK degraders in clinical trials [abstract]. In: Proceedings of the Fifth AACR International Meeting on Advances in Malignant Lymphoma: From Discovery to Clinical Impact; 2026 Jun 24-27; Philadelphia, PA. Philadelphia (PA): AACR; Blood Cancer Discov 2026;7(3_Suppl):Abstract nr PR004.

Abstract Large Language Models (LLMs) demonstrate strong capabilities in healthcare applications. However, recruiting human evaluators to assess usability and effectiveness becomes challenging in the large-scale LLM-generated outcomes. Existing metrics are often outdated for evaluating LLM-based healthcare technologies. This paper proposes a Research through Evaluation (RtE) approach that can refine metrics and improve assessments of LLM-generated outputs. We deploy RtE in a retrospective comparative study of prostate cancer patient inquiries, comparing responses from human clinical care teams with those generated by a GPT-4-based in-basket bot. Through three rounds of co-evaluations with clinical professionals and an LLM-as-graders study with four LLMs, the RtE method shows that starting with small, iterative evaluations helps human graders adjust metrics to suit their daily practices better. Additionally, in the LLM-as-graders study, LLMs exhibit similar behavior, preferring the in-basket bot's responses over the clinical care team. Our methodology and findings demonstrate that the RtE method can enhance the transparency and robustness of evaluation metrics for LLM-generated content, with the potential to generalize to evaluations in other clinical practices.

e15509 Background: Large language models (LLMs) are increasingly explored for oncology decision support, yet comparative evidence evaluating medical-specific versus general-purpose LLMs using expert-reviewed clinical cases is limited. We compared a medical-specific LLM (Google MedGemma 27B) with a general-purpose LLM (ChatGPT-5) using gastrointestinal (GI) and genitourinary (GU) oncology vignettes. Methods: This cross-sectional study evaluated 32 de-identified oncology vignettes (26 GI, 6 GU) created by a U.S. board-certified medical oncologist and informed by real-world cases. Vignettes included text-only information on the clinical context, description of pathology, tumor staging, biomarkers, and relevant comorbidities. Both LLMs were prompted to provide National Comprehensive Cancer Network(NCCN) guideline-based treatment recommendations, monitoring strategies, maintenance therapy considerations, and references, along with a self-reported confidence score (0–100). Board-certified U.S. medical oncologists independently reviewed outputs using a structured instrument; two disease specific expert oncologists evaluated each case. Outcomes included correctness and comprehensiveness (1–10 scales), hallucinations (factually fabricated or unsupported statements), NCCN traceability, potential patient harm, and clinician trust. Associations between accuracy and comprehensiveness were assessed using correlation coefficients. Inter-rater reliability was assessed using a two-way random-effects intraclass correlation coefficient (ICC). Results: ChatGPT-5 demonstrated higher mean correctness (8.65 vs 4.68) and comprehensiveness (8.68 vs 5.22) scores compared with MedGemma 27B. Accuracy and comprehensiveness were strongly correlated overall (Pearson r = 0.93, p < 0.001), with significant model-specific correlations (ChatGPT-5 r = 0.86, p < 0.001; MedGemma r = 0.77, p < 0.001). Hallucinations occurred less frequently with ChatGPT-5 (12.3% vs 65%), and NCCN-traceable recommendations were more common (75.4% vs 36.7%). Responses judged as potentially harmful occurred in 18.5% of ChatGPT-5 outputs versus 70% of MedGemma outputs. Clinician trust favored ChatGPT-5 (84.6% vs 15%). Inter-rater reliability improved with rater averaging (ICC(A,k): correctness 0.72; comprehensiveness 0.64). Conclusions: In expert-reviewed GI and GU oncology cases, ChatGPT-5 demonstrated higher clinical accuracy, comprehensiveness, and guideline alignment than a medical-specific LLM; however, a substantial rate of hallucinations and potentially harmful recommendations persisted across models. These findings underscore that, despite performance differences, current LLMs require rigorous clinician oversight and should not be used as standalone decision-support tools in oncology.