Toesca, Diego A.S.

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.

Abstract Purpose Lattice radiation therapy (LRT) is a form of spatially fractionated radiation therapy that allows increased total dose delivery aiming for improved treatment response without an increase in toxicities, commonly utilized for palliation of bulky tumors. The LRT treatment planning process is complex, while eligible patients often have an urgent need for expedited treatment start. In this study, we aimed to develop a simulation‐free workflow for volumetric modulated arc therapy (VMAT)‐based LRT planning via deep learning‐predicted synthetic CT (sCT) to expedite treatment initiation. Methods Two deep learning models were initially trained using 3D U‐Net architecture to generate sCT from diagnostic CTs (dCT) of the thoracic and abdomen regions using a training dataset of 50 patients. The models were then tested on an independent dataset of 15 patients using image similarity analysis assessing mean absolute error (MAE) and structural similarity index measure (SSIM) as metrics. VMAT‐based LRT plans were generated based on sCT and recalculated on the planning CT (pCT) for dosimetric accuracy comparison. Differences in dose volume histogram (DVH) metrics between pCT and sCT plans were assessed using the Wilcoxon signed‐rank test. Results The final sCT prediction model demonstrated high image similarity to pCT, with a MAE and SSIM of 38.93 ± 14.79 Hounsfield Units (HU) and 0.92 ± 0.05 for the thoracic region, and 73.60 ± 22.90 HU and 0.90 ± 0.03 for the abdominal region, respectively. There were no statistically significant differences between sCT and pCT plans in terms of organ‐at‐risk and target volume DVH parameters, including maximum dose (Dmax), mean dose (Dmean), dose delivered to 90% (D90%) and 50% (D50%) of target volume, except for minimum dose (Dmin) and (D10%). Conclusion With demonstrated high image similarity and adequate dose agreement between sCT and pCT, our study is a proof‐of‐concept for using deep learning predicted sCT for a simulation‐free treatment planning workflow for VMAT‐based LRT.

e16457 Background: Patients with inoperable stage II-III pancreatic cancer commonly undergo sequential stereotactic body radiation therapy (SBRT) and chemotherapy. However, there is variability in clinical outcomes, and the predictors of high-risk patients with rapid tumor progression (within 3 months) and poor overall survival (OS) despite treatment are not well characterized. We developed a radiomic (imaging-derived features) signature (RS) that predicts rapid progression. We then investigated clinical characteristics, including pathologic features and treatments received, and RS in a combined analysis to develop and validate the highest-performing model to predict OS from the time of SBRT initiation. We identified the most predictive feature from the combined model that best predicts OS. Methods: In our retrospective study, we examined a cohort of 124 stage II-III pancreatic cancer patients who underwent sequential SBRT and systemic chemotherapy and had pre-treatment pancreatic protocol computed tomography (CT) imaging. We examined 10 clinical features and extracted 900 radiomic features from each segmented tumor per patient using PyRadiomics. Dividing our cohort into training and test sets (60:40), we built a prediction model for rapid tumor progression using radiomic data on the training set (n = 74), applying a LASSO-based algorithm for feature selection and 5-fold cross-validation for parameter optimization. We validated the model performance on the held-out test set (n = 50) and generated the RS for predicting rapid progression. We examined the performance of clinical features and RS in predicting OS in univariate and multivariate Cox regression models. Results: Analysis on our cohort (57 men, 67 women; mean age 67 ± 11 years) generated a 43-feature RS that predicted rapid tumor progression (AUC 0.83, 95% CI: 0.70–0.94) in the test set. High RS was a significant predictor of mortality with hazard ratio (HR) 2.22 (95% CI: 1.32–3.73, p = 0.003) within the first year and 2.85 (95% CI: 1.35–6.03, p = 0.006) thereafter. Non-intensive chemotherapy increased only early mortality risk (HR 1.95, 95% CI: 1.08–3.53, p = 0.03), while older age was significant in later years (HR 1.80, 95% CI: 1.02–3.15, p = 0.04). Conclusions: CT-derived RS accurately predicted rapid tumor progression in stage II-III pancreatic cancer patients undergoing SBRT sequentially with chemotherapy. High RS was the strongest prognostic indicator for increased mortality risk, suggesting its utility for guiding treatment or selection for clinical trials.

Abstract Background Inter‐fractional and intra‐fractional positioning and motion monitoring based on implanted fiducials through kilovoltage (kV) imaging is a cost‐effective approach to enhance intact prostate radiotherapy treatment accuracy. However, comprehensive studies for three‐dimensional (3D) position extraction and the corrective action threshold are lacking. Purpose To develop and verify a fiducial motion monitoring system with 3D position information based on triggered kV images taken during treatment delivery, and to comprehensively evaluate the recommended thresholds for corrective action during treatment. Methods An in‐house fiducial triangulation algorithm was developed to monitor fiducial positions using subsequent kV triggered images. The precision of fiducial triangulation and motion detection was validated on a pelvis phantom with four fiducial inserts. A retrospective study was conducted on prostate cancer patients who received either moderately hypofractionated radiotherapy (MHRT, n = 16) or stereotactic body radiotherapy (SBRT, n = 12), categorized by endo‐rectal balloon (ERB) use. Intra‐fractional positions relative to isocenter were computed, and these positions were analyzed in comparison to the original marker locations of the patient's treatment plan. A linear regression fit was used per fraction to determine correlation coefficients between motions in the left–right (LR), superior–inferior (SI), and anterior–posterior (AP) directions. Average fraction time (AFT) of treatment was reported based on the observed average time spent on steps in the workflow and the percentage of fiducials caught outside of the current two‐dimensional (2D) tolerance threshold of 5 mm. The observed 3D offsets were used to estimate AFT for various 3D tolerance thresholds. Results Overall, our triangulation method proved to be very precise for static cases, where phantom measurements revealed a fiducial position triangulation precision of mm for stationary targets, but had a spread of mm for targets with 1 mm of motion. Employing triangulation, true motion exceeding 5 mm was detected above 4.0 mm in magnitude 90.9% of the time, a noticeable improvement in comparison to the 28%–65% successful detection rate reported by 2D projection methods. Most detection errors were attributed to depth disparity. Similar to other reports, correlation coefficients for intra‐fractional motion generally indicated no LR/SI, no LR/AP, and weak positive SI/AP correlations for MHRT and SBRT patients. Offsets beyond a 3D tolerance threshold of 5 mm were observed at a rate of 4.25%–5.25%, while the 5 mm 2D out‐of‐tolerance catch‐rate was 1.6%. The AFT was 8.1–8.2 min using the 2D tolerance threshold of 5 mm. In comparison, the estimated AFT for the proposed 3D tolerance monitoring of offsets beyond 2–5 mm was slightly higher at 8.2–10.9 min due to the higher amount of out‐of‐tolerance instances for the higher precision intra‐fractional motion management. Conclusions Our study showcases a promising subsequent kV‐based triangulation method for intra‐fractional prostate motion monitoring. Acquiring 3D motion information results in higher out‐of‐tolerance catch‐rates particularly in the depth dimension of the kV images, which is perpendicular to the treatment beam. Failure to properly observe and catch these offsets would result in sub‐optimal conformality and accuracy of the dose delivery.

Abstract Background Anal squamous cell carcinoma is a rare cancer with increased occurrence of multiple cancers before and after the anal squamous cell carcinoma diagnosis. However, there are limited data on this aspect. This multi-institutional analysis aimed to define the occurrence of malignancies before and after anal squamous cell carcinoma, time trends, and impact on survival and to identify prognostic factors. Methods Initial primary malignancy was defined as a malignancy occurring before the anal squamous cell carcinoma. Second primary malignancy was defined as a distinct primary cancer that developed after anal squamous cell carcinoma diagnosis. Retrospective multi-institutional chart review was done. Progression-free survival (PFS), overall survival, and prognostic factors were evaluated. Results A total of 647 patients with anal squamous cell carcinoma treated with curative intent were analyzed. Median age was 61.2 years with 72% as females. Of these, 150 (23.3%) patients had multiple malignancies with initial primary malignancy in 16% and second primary malignancy in 8%. Patients without prior cancer had better 5-year PFS (81.2% vs 67.2%, P = .011) and overall survival (81% vs 69%, P = .008) compared with those with prior cancer. Second primary malignancies had a statistically significant adverse impact on PFS (hazard ratio [HR] = 4.22) and overall survival (HR = 3.56). Females had better 5-year PFS (82% vs 70%, P = .016) as compared with males. The median time interval for developing anal squamous cell carcinoma (as second primary malignancy) after initial primary malignancy was 9.32 years. Conclusions Anal squamous cell carcinoma patients have an increased risk of multiple malignancies. These patients who have prior cancers have inferior outcomes. Second primary malignancy is a poor prognostic factor in patients with anal cancer. Second primary malignancy can develop years after treatment of primary anal squamous cell carcinoma.

74 Background: We identified computed tomography (CT)-derived radiomic features predictive of tumor progression within three months, then examined their ability to prognosticate overall survival (OS) along with clinical features in pancreatic cancer. We evaluated these features in patients with unresected pancreatic cancer who underwent stereotactic body radiation therapy (SBRT) in sequence with chemotherapy, but not surgery. Methods: In this retrospective study, we examined a cohort of 101 patients with stage II-III pancreatic cancer who underwent SBRT with sequential chemotherapy at a single institution (Stanford Health Care) between 1999-2020. From their pre-SBRT contrast-enhanced CT images with segmented tumors, delineating regions-of-interest, we extracted 900 radiomic (quantitative pixel-level imaging characteristic) features. In the first phase, we identified radiomic features that predicted rapid tumor progression within three months following SBRT. We divided the dataset into a training set (n = 53) for model development and a test set (n = 48) for evaluation. Using logistic regression with the Least Absolute Shrinkage and Selection Operator algorithm for feature selection and classification, we built a binary prediction model on the training set to identify patients at risk of progression within three months of SBRT. To fine-tune parameters, we performed five-fold cross-validation (CV) on the training set, repeating each set of parameters five times. We assessed model performance on the test set using the area under the curve (AUC). We selected the model with the best AUC, generating the predictive radiomic feature set. In the second phase, we conducted univariate and multivariate Cox regression analyses to assess the relationship between OS and individual clinical variables (age, sex, stage, vessel involvement, tumor location, performance status, body mass index, biological equivalent dose of radiation) and the radiomic feature set as high versus low risk. Results: Our cohort consisted of 48 men (mean age, 70 years ± 11 [SD]) and 53 women (mean age, 67 years ± 13 [SD]). From the first phase, 32 textural features comprised the radiomic feature set that best predicted rapid tumor progression, with mean AUCs of 0.852 (CV, n=53) and 0.814 (test, n=48). In the univariate Cox model, only the radiomic feature set was predictive of OS (hazard ratio, HR, 1.724, p=0.011). In the multivariate Cox model, radiomic features and age were significant predictors of OS, with HR of 1.819 (p=0.007) and 1.024 (p=0.024), respectively. Conclusions: CT-derived radiomic features predict rapid tumor progression following SBRT, confer nearly a twofold increase in mortality risk, and, along with patient age, enhance the identification of patients with stage II-III pancreatic cancer with poor OS. [Table: see text]

56 Background: We sought to identify the clinical determinants of overall survival (OS) in patients diagnosed with stage II-III pancreatic cancer not undergoing surgery. We curated a relatively homogenous cohort of patients who did not undergo surgical resection, but instead underwent chemotherapy in sequence with stereotactic body radiation therapy (SBRT) to examine the features that differentiated those with poor OS. Methods: In a retrospective study, we examined a cohort of 259 patients with stage II-III pancreatic cancer who underwent chemotherapy in sequence with SBRT without surgery at a single institution (Stanford Health Care). We conducted both univariate and multivariate Cox regression analyses to assess the relationships between OS and individual clinical and tumoral variables, including: age, sex, stage, Karnofsky performance status (KPS), body mass index, vessel involvement, tumor location, and biological equivalent dose [BED] of SBRT as potential determinants. Results: The study cohort consisted of 136 men (52.5%) and 123 women (47.5%) with a median age of 69 (range 38-94). Tumor histology was predominantly ductal adenocarcinoma (n=249, 96.1%), with tumor sites of involvement spanning: uncinate (n=35, 13.5%), head (n=160, 61.8%), body (n=59, 22.8%), and tail (n=5,1.9%). More patients had stage III (n=171, 66%) than stage II (n=88, 34%) disease, and their tumor resectability by NCCN criteria were rated as: unresectable (n=145, 56%), borderline (n=28, 10.8%), resectable (n=11, 4.2%), or unrecorded (n=75, 29%). Tumor progression was observed in local (n=71, 27.4%), regional (n=29, 11.2%), and distant (n=150, 57.9%) sites during the median follow up period of 17 months (range 15-18 months) from diagnosis and 12 months (10-13 months) from SBRT. Death (n=213, 82.2%) during this period was more common than survival (n=46, 17.8%). In univariate analyses, clinical features impacting OS included: KPS (hazard ratio [HR] 0.14, p=0.007); male sex (HR 1.44, p=0.01), pancreatic head tumors (HR 1.44, p=0.011), BED (HR 1.33, p=0.035), and age (HR 1.013, p=0.044). In the multivariate analysis, which incorporated only features significant in the univariate analyses, tumor location at the pancreatic head (HR 1.412, p=0.021) and male sex (HR 1.360, p=0.03) conferred the highest risks for poor OS, with BED bordering on significance (p=0.059) (Table). Conclusions: Among patients with stage II-III pancreatic cancer who underwent chemotherapy in sequence with SBRT, but not surgery, the most significant clinical determinants of poor OS were location of the tumor in the pancreatic head and male sex. Prioritizing additional disease monitoring and alternative therapies for this subset of patients at higher risk of poor survival may be warranted. [Table: see text]

AbstractPurposeTo assess the practicality of employing a commercial knowledge‐based planning tool (RapidPlan) to generate adapted intact prostate and prostate bed volumetric modulated arc therapy (VMAT) plans on iterative cone‐beam computed tomography (iCBCT) datasets.Methods and materialsIntact prostate and prostate bed RapidPlan models were trained utilizing planning data from 50 and 44 clinical cases, respectively. To ensure that refined models were capable of producing adequate clinical plans with a single optimization, models were tested with 50 clinical planning CT datasets by comparing dose‐volume histogram (DVH) and plan quality metric (PQM) values between clinical and RapidPlan‐generated plans. The RapidPlan tool was then used to retrospectively generate adapted VMAT plans on daily iCBCT images for 20 intact prostate and 15 prostate bed cases. As before, DVH and PQM metrics were utilized to dosimetrically compare scheduled (iCBCT Verify) and adapted (iCBCT RapidPlan) plans. Timing data was collected to further evaluate the feasibility of integrating this approach within an online adaptive radiotherapy workflow.ResultsModel testing results confirmed the models were capable of producing VMAT plans within a single optimization that were overall improved upon or dosimetrically comparable to original clinical plans. Direct application of RapidPlan on iCBCT datasets produced satisfactory intact prostate and prostate bed plans with generally improved target volume coverage/conformality and rectal sparing relative to iCBCT Verify plans as indicated by DVH values, though bladder metrics were marginally increased on average. Average PQM values for iCBCT RapidPlans were significantly improved compared to iCBCT Verify plans. The average time required [in mm:ss] to generate adapted plans was 06:09 ± 02:06 (intact) and 07:12 ± 01:04 (bed).ConclusionThis study demonstrated the feasibility of leveraging RapidPlan to expeditiously generate adapted VMAT intact prostate and prostate bed plans on iCBCT datasets. In general, adapted plans were dosimetrically improved relative to scheduled plans, emphasizing the practicality of the proposed approach.

706 Background: We sought to assess the ability of a computed tomography (CT)-derived radiomic signature (RS) feature set to predict rapid tumor progression and overall survival (OS) in patients diagnosed with stage II-III pancreatic cancer undergoing stereotactic body radiation therapy (SBRT) in sequence with chemotherapy. Methods: We conducted a retrospective study on a cohort of 240 stage II-III pancreatic cancer patients who underwent SBRT in sequence with chemotherapy at a single institution (Stanford Hospital and Clinics). Among them, 83 had pre-SBRT contrast-enhanced CT images with segmented tumors, forming the imaging set, from which we extracted 900 radiomic (quantitative pixel-level imaging characteristic) features from delineated tumor regions-of-interest. For RS development, we divided the imaging set into a training set (n = 53) for model development and a test set (n = 30) for evaluation. We built a binary prediction model on the training set to identify patients at risk of rapid tumor progression within three months after SBRT. We utilized logistic regression with the Least Absolute Shrinkage and Selection Operator algorithm for feature selection and classification. To fine-tune parameters, we performed five-fold cross-validation on the training set, repeating each set of parameters five times. Finally, we assessed the model's performance on the test set (n = 30) using the area under the curve (AUC) value. We selected the model with the best AUC, while also generating the predictive radiomic feature set. For OS prognostication, we conducted both univariate and multivariate Cox proportional-hazards analyses using RS and clinical features (age, sex, stage, vessel involvement, tumor location, performance status, body mass index, biological equivalent dose of radiation) as potential predictors. Results: The enrolled cohort consisted of 127 men (mean age, 69 years ± 12 [SD]) and 113 women (mean age, 69 years ± 13 [SD]). Seventeen textural radiomic features were identified as the RS, which demonstrated a high AUC in the test set for the prediction of rapid progression (AUC 0.850; 95% CI: 0.725, 0.975). Age was associated with OS in the multivariate model while high RS (>0.5) was independently associated with OS in both univariate and multivariate models. Conclusions: CT-derived RS, combined with age, represent the most prognostic factors for stage II-III pancreatic cancer. RS, which likely reflects underlying biology or molecular alterations, contributes to the highest HR of poor OS and provide the strongest indicator of outcome following commonly used treatments in this patient population. [Table: see text]