Laughlin, Brady

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 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 Minimal clinically important differences (MCIDs) quantify the clinical relevance of quality of life results at the individual patient and group level. The aim of this study was to estimate the MCID for the Brief Fatigue Inventory (BFI) and the Worst and Usual Fatigue items in patients with brain or CNS cancer undergoing curative radiotherapy. Methods Data from a multi-site prospective registry was used. The MCID was calculated using distribution-based and anchor-based approaches. For the anchor-based approach, the fatigue item from the PROMIS-10 served as the anchor to determine if a patient improved, deteriorated, or had no change from baseline to end of treatment (EOT). We compared the unadjusted means on the BFI for the 3 groups to calculate the MCID. For the distribution-based approaches, we calculated the MCID as 0.5 SD of the scores and as 1.96 times the standard error of measurement. Results Three-hundred and fifty nine patients with brain or CNS tumors undergoing curative radiotherapy filled out the 9-item BFI at baseline and EOT. The MCID for the BFI was 1.33 (ranging from 0.99 to 1.70 across the approaches), 1.51 (ranging from 1.16 to 2.02) and 1.76 (ranging from 1.38 to 2.14) for the usual and worst fatigue items, respectively. Conclusions This study provides the MCID ranges for the BFI and Worst and Usual fatigue items, which will allow clinically meaningful conclusions to be drawn from BFI scores. These results can be used to select optimal treatments for patients with brain or CNS cancer or to interpret BFI scores from clinical trials.

Abstract PURPOSE 18F-DOPA is sensitive/specific for identifying areas of biologically aggressive disease for glioblastoma. We report the impact of 18F-DOPA on radiation treatment planning from a phase II study. METHODS Gross target volume (GTV) was defined by the boolean of high-risk PET (tumor/normal brain SUV > 2.0) and MRI T1 contrast enhancement (MRI-CE) including surgical cavity. The clinical target volume (CTV) included the GTV with 1cm margin. Using proton beam therapy, GTV ≤ 65 cc was treated with 25 (CTV), 30 (MRI-CE), and 35 (PET) GyE over 5 fractions with a simultaneous integrated boost. GTV > 65 cc received 30 (CTV), 35 (MRI-CE), and 40 (PET) GyE over 10 fractions. Radiation field design was compared to traditional 1.5cm CTV margin on MRI-CE. RESULTS Between May 2019 and October 2021, 39 patients were treated. The median MRI-CE tumor diameter was 5.2cm. The median PET volume was 7.9cc (range 0–64.9cc). The median MRI-CE volume was 35.1cc (range 8.6–114cc). High-risk PET was contained within: MRI-CE 18/39 (46%), MRI-CE + T2W FLAIR 13/39 (33%), MRI-CE + T2W FLAIR + outside GTV 4/39 (10%), outside GTV 2/39 (5%) and no uptake 2/39 (5%). Utilizing this planning technique, there was a 26% volume reduction in CTV to traditional 1.5 cm margin on MRI-CE. PET imaging in treatment planning allowed inclusion of high-risk disease that would have been excluded on traditional CTV in 23% of cases (range 1.6-33.8cc). PET was more predictive [HR 1.03 (95% CI 1.01, 1.05) p = 0.002] than MRI [HR 1.01 (95% CI 1.00, 1.03) p = 0.02] for death with each cc increasing risk of death by 3.1%. CONCLUSIONS The incorporation of 18F-DOPA in treatment planning yields a more targeted approach with reductions in both normal brain treated and risk of geographic miss compared to traditional CTV margins.

494 Background: Our aim was to compare survival between patients receiving neoadjuvant chemoradiation and orthotopic liver transplantation (OLT group) versus definitive chemoradiation (CRT group) for extrahepatic or hilar cholangiocarcinoma. Methods: 49 patients (20 in OLT group vs. 29 in CRT group) with unresectable hilar/extrahepatic cholangiocarcinoma were treated at Mayo Clinic Arizona between Feb. 1998–Sep. 2019. Treatment included external beam radiation therapy (median 4500cGy) and boost (median 900cGy) with either continuous 5-flurouracil (dose range 180–225 mg/m2) or capecitabine (dose range 825–1000 mg/m2 BID) prior to or without OLT. Radiation boosts were delivered with EBRT or bile duct brachytherapy. Patients were between 27.9–84.3 years (median 64.3) at diagnosis. 18 patients had previous diagnosis of PSC. Results: Between Feb. 1998–Sep. 2019, 31(63%) of 49 patients died by the end of follow-up. Of patients treated with neoadjuvant therapy and OLT, 7(35%) of 20 patients died. 24(86%) of 28 patients treated with definitive therapy died. The OLT cohort were younger (mean age 56.5 vs. 69.0 years), more likely to have PSC and UC (65% vs. 17%), and had a lower CA 19-9 (median 43 vs. 535)(P < 0.003). From the end date of radiation, median overall survival was 76.8 months vs. 15.6 months for the OLT and CRT groups, respectively. Survival rates at 3 and 5 years were 78% and 69% in the OLT group compared to 19% and 6% in the CRT group (HR 7.73; 3.04-19.65:(P < 0.0001)). Progression-free survival (89% vs. 30% at 3 years), and distant metastasis-free survival (88% vs. 66% at 3 years) favored OLT versus CRT alone (HR 5.74;1.12-29.34:(P < 0.02)). Univariate analysis demonstrated that the method of treatment (OLT vs. CRT) was the only variable associated with better clinical outcomes. Conclusions: In patients with unresectable extrahepatic/hilar cholangiocarcinoma, survival was higher in those who underwent chemoradiation and OLT. Patients who received definitive chemoradiation in the absence of OLT were expected to have worse overall, progression-free, and metastasis-free survival.

Abstract Background: Breast tumors exhibit intratumor heterogeneity resulting in targeted therapy resistance and other challenges in disease management. To address the sources of heterogeneity, we performed a unique, in-depth analysis of clonal architecture in primary chemoradiation-naïve breast cancers. We combined DNA content-based flow cytometry and ploidy analysis with aCGH and whole exome next-generation sequencing (NGS) in multiple biopsies from the tumors and involved lymph nodes (LNs). Material and methods: We sorted nuclei from distinct populations of diploid, tetraploid, and aneuploid cells in surgical tumor samples from three chemoradiation-naïve patients. Each sorted tumor cell population was interrogated with aCGH and whole exome NGS. In Patient #1, we sorted and interrogated the genomes of tumor populations from 12 fresh frozen sections morphologically mapped from within a HER2+, ER+, PR- primary invasive ductal carcinoma (IDC) of histological grade 3 with LN involvement and 2-3 sections from 2 out of 5 LNs. In Patient #2, 11 morphologically mapped fresh frozen sections were analyzed from a grade 2, ER+, PR+, HER2+, BRCA2 mutant LN- IDC. In parallel, matching samples were processed for IHC assays. In Patient #3 biopsies from a grade 2 ER+, PR+ primary tumor were profiled. Results: The 18 primary and LN biopsies from Patient #1 fell into 6 distinct ploidy groups albeit with aberrant but homogenous aCGH profiles, characterized by SARC amplification and homozygous deletions in ROBO1 and ROBO2. In contrast a dominant ploidy was identified throughout 10 biopsies in Patient #2 but with heterogeneous aCGH profiles. Patient #3 had a clonal homozygous deletion in Numb in each of 4 tumor biopsies. Mutation profiles obtained through exome sequencing further confirmed that ploidy was the main driver in Patient #1 whereas copy number aberrations played the key role in Patient #2 with the BRCA2 mutation (R3129X). A dendrogram based on exome variant calls of the aneuploid populations in Patient #1 strongly correlated with ploidy group and further revealed the specific clonal population characterized by a 5N ploidy and homozygous mutations in TP53 and PIK3CA as the progenitor to the ploidies present in the distant LNs. Strikingly, patients #1 and #2 were HER2 wild type across their clonal populations, contradicting IHC staining in a single core biopsy. Conclusions:Rather than inferring the presence of distinct tumor cell populations, our flow-sorting based approach of first identifying clonal populations and then interrogating their genomes, provides an objective method of exploring the sources and clinical significance of tumor heterogeneity. Our approach of clonal analysis has broad implications in the study of tumor heterogeneity and the identification of drivers in breast and other solid tumors that can advance more effective treatment and clinical management of patients with this disease. Citation Format: MARIACARLA ANDREOZZI, Princy Francis, Elizabeth Lenkiewicz, Mia Champion, Brady Laughlin, Karen Anderson, Heather Cunliffe, Ann E. McCullough, Michael T. Barrett, Barbara Pockaj. Deep clonal profiling identifies distinct mechanisms of heterogeneity and evolution in breast cancer. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr LB-175. doi:10.1158/1538-7445.AM2015-LB-175