Vora, Sujay A.

Abstract Background Leptomeningeal disease (LMD) remains a challenging complication of both solid tumors and hematologic malignancies, with diagnosis traditionally relying on cerebrospinal fluid (CSF) cytology, which has limited sensitivity. Circulating tumor DNA (ctDNA) analysis in CSF has emerged as a promising diagnostic and molecular profiling tool that may improve detection and guide treatment decisions. We will report real-world clinical experience using a CSF ctDNA assay (Menarini Silicon Biosystems) in routine neuro-oncology practice. Methods Retrospective analysis of over 100 CSF specimens obtained from patients with suspected or confirmed LMD will be conducted. Clinical data to be collected: patient demographics, tumor histology, indication for testing, and prior treatments. ctDNA assay results will be compared with conventional CSF cytology. Molecular alterations identified through ctDNA testing will be recorded, along with subsequent treatment decisions, including use of targeted therapies, systemic therapy modifications, radiation therapy, or intrathecal approaches. Preliminary clinical outcomes, including radiographic response, progression patterns, and survival trends, will be explored. Descriptive statistical methods will be used to summarize the findings. Results We anticipate that among analyzed specimens, ctDNA testing will provide clinically meaningful diagnostic yield across multiple tumor histologies. Concordance and discordance patterns between ctDNA results and conventional cytology will be evaluated, including cases in which ctDNA identified molecular alterations despite negative cytology. Actionable genomic findings in a subset of patients will be analyzed and presented. Early outcome observations that ctDNA-informed management may facilitate more personalized care will be reported. Conclusions CSF ctDNA analysis represents a feasible and informative adjunct to conventional cytology in the evaluation of patients with suspected LMD in real-world clinical practice. Beyond diagnostic support, molecular profiling from ctDNA may influence treatment selection and enable precision oncology approaches in this high-risk population. Prospective studies are warranted to define sensitivity, specificity, and the impact of ctDNA-guided interventions on clinical outcomes.

Abstract Background The current treatment of glioblastoma involves resection followed by radiotherapy (RT) and temozolomide. Rapid early progression (REP) of tumor in the surgery-to-radiotherapy interval (SRI) has been described using categorical or semi-quantitative assessments and is associated with inferior clinical outcomes. This study seeks to establish and validate a quantitative definition for REP and examine how REP could inform future therapy. Methods Volumes of tumor and resection cavity at 3 time points—preoperative, postoperative and RT-planning (pre-RT)—were contoured among patients who had undergone standard of care. A minimal, but clinically appreciable, threshold for defining REP based on change in tumor volume was established, validated, and subsequently used to assess actionable predictors of REP and its prognostic value. Results An increase of >2 cc in volume or a new satellite tumor between the post-operative to planning MRI was found to be a sensitive, pragmatic, and statistically significant cutoff for REP (P = .002). Actionable predictors such as SRI of 6 weeks and a > 5 cc residual immediate post-operative tumor volume were found to be statistically significant predictors of REP. REP was also found to be a statistically significant prognostic indicator for overall survival (P = .002) and progression-free survival. Conclusion These findings support that an increase in tumor volume between post-operative and pre-RT MRI of >2 cc is the ideal working definition of REP in this clinical setting. Furthermore, REP could serve as a prognostic ­biomarker to track treatment response. Additionally, 2 actionable predictors of REP, post-op residual tumor volume and SRI, are identified.

Abstract Background Accurate and efficient dose calculation is essential for online adaptive planning in proton therapy. Deep learning (DL) has shown promising dose prediction results for pencil beam scanning proton therapy (PBSPT) in recent years, but existing DL‐based dose prediction methods still suffer from limited generalizability and an inability to effectively handle outlier clinical cases. This may lead to inaccurate dose delivery to targets or excessive irradiation to organs at risk (OARs), thereby compromising the safety and efficacy of online adaptive proton therapy. Purpose To design a physics‐aware and generalizable AI‐based PBSPT dose prediction method that incorporates underlying physics to enhance generalizability, particularly in handling outlier clinical cases. Methods This study analyzed PBSPT plans of 103 prostate (93 for training and 10 for testing) and 78 lung cancer patients (68 for training and 10 for testing) from our institution, with each case comprising CT images and structure sets. Using the doses generated by our Monte Carlo‐based dose engine as the reference standard, we compared three methods: the region of interest (ROI)‐based method, the beam mask and sliding window method, and the proposed noisy probing dose method, which rapidly generates a low‐statistics dose via uniformly weighted spots on an expanded spot‐placement target volume without optimization. To evaluate the generalizability of these methods to rare treatment planning scenarios, 12 cases with uncommon beam angles or prescription doses were used to assess their performance, which was evaluated using dose‐volume histogram (DVH) indices, 3D Gamma passing rates (3%/2 mm/10%), and Dice coefficients for dose agreement, while prediction times were measured to gauge model efficiency. Results The proposed noisy probing dose method consistently outperformed the ROI‐based and beam mask baselines across all evaluated metrics, with more accurate dose agreement and superior generalizability. For DVH indices, the noisy probing dose method achieved the smallest deviation in clinical target volume (CTV) dose coverage: in prostate cancer, CTV D98 deviation was reduced by 45% (from 0.53 ± 0.22 Gy [RBE] for ROI‐based) and 29% (from 0.41 ± 0.28 Gy [RBE] for beam mask) to 0.29 ± 0.06 Gy [RBE]; in lung cancer, similar improvements were observed, with CTV D98 deviations reduced to 0.34 ± 0.12 Gy [RBE]. The 3D Gamma passing rates improved to 99.65% ± 1.15% for prostate targets and 97.04% ± 1.17% for lung targets. The dice coefficients of the 90% iso‐dose lines were also the highest with the noisy probing dose method (prostate: 0.983 ± 0.005; lung: 0.967 ± 0.01). For the 12 outlier cases, the noisy probing dose method maintained superior generalizability, yielding higher 3D Gamma passing rates (prostate targets: 96.79% ± 0.83%, OARs: 94.29% ± 1.01%; lung targets: 93.38% ± 1.34%, OARs: 93.95% ± 1.32%), demonstrating robust generalizability to rare clinical scenarios. The dose predictions for all testing cases were completed within 0.3 seconds. Conclusions A novel noisy probing dose method was proposed for PBSPT dose prediction in prostate and lung cancer patients. By embedding more proton‐specific physics, this method demonstrated an improvement in the generalizability of dose prediction.

Abstract Background In the current clinical workflow of radiation oncology, therapists manually summarize the physician‐issued Computed Tomography (CT) simulation orders to prepare patients' CT simulations. This process increases the workload, introduces variability in documentation quality, and is prone to human errors. Purpose This study aims to address these challenges using a large language model (LLM) to automate the generation of summaries from the CT simulation orders and evaluate its performance. Methods A total of 607 CT simulation orders were collected from the Aria database at our institution. A locally hosted LLaMa 3.1 405B model, accessed via the Application Programming Interface (API) service, was used to extract keywords from the CT simulation orders and generate summaries. The downloaded CT simulation orders were categorized into seven groups based on treatment modalities and disease sites. For each group, a customized instruction prompt was developed collaboratively with therapists to guide the LLaMa 3.1 405B model in generating summaries. The ground truth for the corresponding summaries was manually derived by carefully reviewing each CT simulation order and subsequently verified by therapists. The accuracy of the LLM‐generated summaries was evaluated by therapists using the ground truth. Results Over 98% of the LLM‐generated summaries aligned with the ground truth in terms of accuracy. Our evaluations showed an improved consistency in format and enhanced readability from the LLM‐generated summaries compared to the corresponding therapist‐generated summaries. This automated approach demonstrated consistent performance across all groups, regardless of treatment modality or disease site. Conclusions This study demonstrated the high precision and consistency of the LLaMa 3.1 405B model in extracting keywords and summarizing CT simulation orders, suggesting that LLMs have great potential to assist in this task, reduce the workload of CT simulation therapists and improve radiation oncology workflow efficiency.

Abstract Background The radiation oncology clinical practice involves many steps relying on the dynamic interplay of abundant text data. Large language models have displayed remarkable capabilities in processing complex text information. But their direct applications in specific fields like radiation oncology remain underexplored. Purpose This study aims to investigate whether fine‐tuning LLMs with domain knowledge can improve the performance on Task (1) treatment regimen generation, Task (2) treatment modality selection (photon, proton, electron, or brachytherapy), and Task (3) ICD‐10 code prediction in radiation oncology. Methods Data for 15 724 patient cases were extracted. Cases where patients had a single diagnostic record, and a clearly identifiable primary treatment plan were selected for preprocessing and manual annotation to have 7903 cases of the patient diagnosis, treatment plan, treatment modality, and ICD‐10 code. Each case was used to construct a pair consisting of patient diagnostics details and an answer (treatment regimen, treatment modality, or ICD‐10 code, respectively) for the supervised fine‐tuning of these three tasks. Open source LLaMA2‐7B and Mistral‐7B models were utilized for the fine‐tuning with the Low‐Rank Approximations method. Accuracy and ROUGE‐1 score were reported for the fine‐tuned models and original models. Clinical evaluation was performed on Task (1) by radiation oncologists, while precision, recall, and F‐1 score were evaluated for Task (2) and (3). One‐sided Wilcoxon signed‐rank tests were used to statistically analyze the results. Results Fine‐tuned LLMs outperformed original LLMs across all tasks with p value ≤ 0.001. Clinical evaluation demonstrated that over 60% of the fine‐tuned LLMs‐generated treatment regimens were clinically acceptable. Precision, recall, and F1‐score showed improved performance of fine‐tuned LLMs. Conclusion Fine‐tuned LLMs demonstrated statistically significant improvements over original LLMs upon three clinically important tasks in radiation oncology. This study explored the feasibility of applying fine‐tuned LLMs in radiation oncology, inspiring further development of utilizing LLMs to assist with radiation oncology tasks.

Background/Objectives: Adjuvant radiation for gynecologic malignancies often exposes organs at risk (OARs), such as the bone marrow, bowel, rectum, and bladder, to radiation, leading to toxicities that impact treatment tolerance and patient quality of life. Scanning proton beam therapy, particularly with Individual Field Simultaneous Optimization (IFSO), may offer dosimetric and biological advantages over volumetric modulated arc therapy (VMAT). This study evaluates the clinical impact of IFSO-based proton planning in post-operative gynecologic cancer patients. Materials and Methods: Fourteen patients receiving adjuvant proton therapy to 45 Gy in 25 fractions were retrospectively analyzed. Comparison VMAT plans were generated on the same datasets. Dose–volume metrics for key OARs and normal tissue complication probabilities (NTCPs) were compared using paired statistical tests. Robustness evaluations accounted for setup and range uncertainties. Results: Proton plans significantly reduced dose to bone marrow (V10Gy: 58% vs. 86%, p < 0.00001; V20Gy: 47% vs. 58%, p < 0.00001), small bowel (V20Gy: 21% vs. 56%, p < 0.00001), and femoral heads (left femoral head mean: 11Gy vs. 13Gy, p = 0.032; right femoral head mean: 11Gy vs. 13Gy, p = 0.022). NTCP modeling predicted significantly lower rates of bowel urgency (9.4% vs. 3.3%, p < 0.001) and hematologic toxicity (10.2% vs. 4.9%, p < 0.001) with proton therapy. Plans remained robust across uncertainty scenarios. Conclusions: IFSO-based scanning proton therapy provides clinically meaningful sparing of bone marrow and bowel, with the potential to reduce hematologic and gastrointestinal toxicities. These findings support its use in patients receiving adjuvant pelvic radiotherapy, particularly those undergoing extended field treatment or chemotherapy.

Objective: We aimed to analyze the current literature for IORT in gynecological cancers and summarized clinical outcomes regarding patient selection. Methods: A systematic search was conducted utilizing PUBMED, Embase, and CINAHL to identify studies following PRISMA-ScR guidelines. A PICOS structure was utilized: population: patients with epithelial gynecological cancers; intervention: IORT; C: a comparator was not required, as we aimed to analyze patient selection; outcome: clinical outcomes and overall survival; and S: experimental and quasi-experimental analytical observational studies and descriptive observational studies, excluding case series published in English and limited to the last 10 years. Data extraction was conducted for patient selection, IORT, oncological outcomes, and morbidity. Results: A total of 707 results were identified, and 509 studies were uploaded to Covidence for screening after removing duplications. Of the 21 eligible studies, 9 were included in the final review. The total number of patients included was 348. The studies were retrospective single-institution studies, except for one. There was significant heterogeneity in their design and protocols. IORT was exclusively used for recurrent and advanced stage gynecological cancers adjunct to pelvic exenteration or laterally extended endopelvic resections with variable indications across institutions. The mean number of IORT patients per study was 2.8 per year. Survival rates were variable and dependent on the surgical margin. Endometrial cancer had a favorable outcome compared to vulvar and cervical cancers. Conclusions: Current clinical practice, as demonstrated by the research, is consistent with NCCN guidelines that endorse the application of IORT in instances of recurrent cervical, vaginal, and vulvar malignancies; however, there are no established recommendations for primary tumors. The analysis shows that there are gaps in our knowledge, mainly regarding the status of the margins, the criteria used to choose patients, and the outcomes that are specific to each histology. The standardization of protocols and prospectively powered studies are needed to refine patient selection criteria.

AbstractBackgroundConventional approaches for emergent or expedited palliative radiotherapy (RT) involve the application of cumbersome vendor‐provided solutions and/or multiple patient appointments to complete the RT workflow within a compressed timeframe.PurposeThis report delineates the clinical development of an in‐house, semi‐automated Cone‐beam computed tomography (CBCT)‐based simulation‐free platform for expedited palliative RT on conventional linacs, intended to supplant existing techniques employed at this institution.MethodsThe internal software, termed SimFree Wizard (SFW), was engineered utilizing a C#‐based application programming interface integrated within the treatment planning system (TPS). Generated scripts were compiled as stand‐alone executables, with a graphical user interface (GUI) customized via an integrated development environment. The platform was conceived as a framework for accelerated CBCT‐based RT, bypassing the requirement for standard simulation imaging. SFW employs full automation where feasible to minimize user intervention, supplemented by graphical instructions for tasks requiring manual execution. During development, relevant temporal metrics from 10 end‐to‐end tests for palliative spine RT were quantified. User feedback was solicited via a simple questionnaire assessing the overall platform usability. Automated in‐house secondary verification software was developed for validation of the TPS‐calculated monitor units (MUs).ResultsThe mean duration for workflow execution was 41:42 ± 3:18 [mm:ss] (range ∼37–46 min). SFW satisfactorily generated simple, multi‐field CBCT‐based 3D treatment plans within seconds following delineation of the desired treatment area. User feedback indicated enhanced usability compared to previously employed solutions. Validation of the secondary verification software demonstrated accurate results for palliative spine RT and other simple cases wherein the dose calculation point resides in a predominantly homogenous medium.ConclusionA novel in‐house solution for expedited CBCT‐based RT was successfully developed, facilitating completion of the entire workflow within approximately 1‐hour or less for simple palliative/emergent scenarios. Overall, this application is expected to improve the quality and safety of palliative RT while greatly reducing workflow duration, thereby improving access to palliative care.

Abstract INTRODUCTION Lymphopenia attributed to postoperative chemoradiotherapy (CRT) can accelerate tumor recurrence and contribute to poor prognosis in GBM. We hypothesize that short course (1-2 weeks) CRT can mitigate lymphopenia compared to standard course and may provide immune benefits for future combination therapy. METHODS GBM patients were randomized on a multi-site phase II study comparing short course (35Gy/5Fx or 40Gy/10Fx depending on target volume) (Arm A) versus standard dose fractionation (60Gy/30Fx or 40Gy/15Fx, Arm B). All patients received concurrent and adjuvant temozolomide (TMZ). This unplanned interim analysis evaluates changes of absolute lymphocyte count (ALC), CD4 and CD8 prospectively at 3-timepoints: baseline (prior to CRT), end of CRT, and 1-month post-CRT. RESULTS To date, 72 patients enrolled (34 Arm A, 38 Arm B). Median age 66 (35-86), ECOG 0-2. Patients characteristics as followed in Arm A vs. Arm B, respectively: female 47% vs. 42%, MGMT promotor methylated 31% vs. 34%, steroid use 46% vs. 45%, ALC 1310 cells/mm3 (330 –9730) vs. 1480 (750 –7750), CD4 531 cells/mm3 (86 –1368) vs. 662 (318 –1671), and CD8 151 cells/mm3 (56 –940) vs. 272 (68 –875). Lymphocytes data were collected in 49 patients (21 Arm A, 28 Arm B). There were signficantly fewer reductions of ALC, CD4 and CD8 in Arm A. Median ALC changes were 95% and 99% of baseline (Arm A) vs 52% and 61% of baseline (Arm B) at the end of CRT and 1-month post-CRT, respectively (p&lt;0.01). Median CD4 changes 93% and 96% of baseline (Arm A) vs 51% and 49% (Arm B), and median CD8 changes 97% and 98% of baseline (Arm A) vs 57% and 71% (Arm B) at the end of CRT and 1-month post-CRT, respectively (p&lt;0.05). CONCLUSIONS This interim analysis suggests that short course RT could reduce treatment-related lymphopenia and immune suppression compared to standard RT, thus could be potentially beneficial to combine with immunotherapy.

Abstract BACKGROUND Fatigue is among the most common symptoms managed by patients with cancer. Hypothalamic pituitary axis dysfunction has an established role in a subset of patients with fatigue. Our primary aim was to longitudinally assess the impact of hypothalamic radiation exposure to patient self-reported fatigue. METHODS We included a sample of patients who received radiation therapy treated at a single institution from 2014-2024. The PROMIS-10 questionnaire (“How would you rate your fatigue on average,” scale 1-5) was used in standard clinical practice, with measures from 6 time points over 2 years. We performed descriptive statistics. Linear mixed models were used to assess the impact of hypothalamic radiation dose on fatigue scores. Mixed models were controlled for gender, age, time, tumor category (CNS vs. non-CNS), and modality (photon vs. proton). Models were fit with random intercepts for each Patient ID. RESULTS 1092 patients (43% female) with a median age of 54 years (range: 18-92) were included. Most patients were treated for CNS cancer (76%), with about 50% each photon and proton. Median hypothalamus radiation dose was 1306 cGy (range: 0.01-6138). Mean baseline fatigue (2.4) increased after treatment (2.7) remaining elevated at 3- and 6-months post-treatment (2.6, 2.5), returning near baseline at 1- and 2-years post-treatment (2.4, 2.3). Higher radiation dose (Estimate (SE): 0.014 (0.006), p=0.0266) and CNS cancer treatment (0.11 (0.051), p=0.0319) were significant predictors of higher fatigue in the mixed model. Male patients (-0.16 (0.042), p=0.0001) and time (-0.003 (0.0016), p=0.0398) were associated with lower fatigue. Age and treatment modality were not significant predictors of fatigue. CONCLUSIONS We found that higher hypothalamic radiation dose was related to more self-reported fatigue, which peaked by 6 months. This effect was seen more prominently in female patients and those treated for CNS disease. Treatment modality and age did not influence fatigue scores.