Integrating local and distant radiation-induced lung injury: Development and validation of a predictive model for ventilation loss.

BACKGROUND: Investigations on radiation-induced lung injury (RILI) have predominantly focused on local effects, primarily those associated with radiation damage to lung parenchyma. However, recent studies from our group and others have revealed that radiation-induced damage to branching serial structures such as airways and vessels may also have a substantial impact on post-radiotherapy (RT) lung function. Furthermore, recent results from multiple functional lung avoidance RT trials, although promising, have demonstrated only modest toxicity reduction, likely because they were primarily focused on dose avoidance to lung parenchyma. These observations emphasize the critical need for predictive dose-response models that effectively incorporate both local and distant RILI effects. PURPOSE: We develop and validate a predictive model for ventilation loss after lung RT. This model, referred to as P+A, integrates local (parenchyma [P]) and distant (central and peripheral airways [A]) radiation-induced damage, modeling partial (narrowing) and complete (collapse) obstruction of airways. METHODS: In an IRB-approved prospective study, pre-RT breath-hold CTs (BHCTs) and pre- and one-year post-RT 4DCTs were acquired from lung cancer patients treated with definitive RT. Up to 13 generations of airways were automatically segmented on the BHCTs using a research virtual bronchoscopy software. Ventilation maps derived from the 4DCT scans were utilized to quantify pre- and post-RT ventilation, serving, respectively, as input data and reference standard (RS) in model validation. To predict ventilation loss solely due to parenchymal damage (referred to as P model), we used a normal tissue complication probability (NTCP) model. Our model used this NTCP-based estimate and predicted additional loss due radiation-induced partial or complete occlusion of individual airways, applying fluid dynamics principles and a refined version of our previously developed airway radiosensitivity model. Predictions of post-RT ventilation were estimated in the sublobar volumes (SLVs) connected to the terminal airways. To validate the model, we conducted a k-fold cross-validation. Model parameters were optimized as the values that provided the lowest root mean square error (RMSE) between predicted post-RT ventilation and the RS for all SLVs in the training data. The performance of the P+A and the P models was evaluated by comparing their respective post-RT ventilation values with the RS predictions. Additional evaluation using various receiver operating characteristic (ROC) metrics was also performed. RESULTS: We extracted a dataset of 560 SLVs from four enrolled patients. Our results demonstrated that the P+A model consistently outperformed the P model, exhibiting RMSEs that were nearly half as low across all patients (13 ± 3 percentile for the P+A model vs. 24 ± 3 percentile for the P model on average). Notably, the P+A model aligned closely with the RS in ventilation loss distributions per lobe, particularly in regions exposed to doses ≥13.5 Gy. The ROC analysis further supported the superior performance of the P+A model compared to the P model in sensitivity (0.98 vs. 0.07), accuracy (0.87 vs. 0.25), and balanced predictions. CONCLUSIONS: These early findings indicate that airway damage is a crucial factor in RILI that should be included in dose-response modeling to enhance predictions of post-RT lung function.

Intricacies of Human-AI Interaction in Dynamic Decision-Making for Precision Oncology: A Case Study in Response-Adaptive Radiotherapy.

Background: Adaptive treatment strategies that can dynamically react to individual cancer progression can provide effective personalized care. Longitudinal multi-omics information, paired with an artificially intelligent clinical decision support system (AI-CDSS) can assist clinicians in determining optimal therapeutic options and treatment adaptations. However, AI-CDSS is not perfectly accurate, as such, clinicians' over/under reliance on AI may lead to unintended consequences, ultimately failing to develop optimal strategies. To investigate such collaborative decision-making process, we conducted a Human-AI interaction case study on response-adaptive radiotherapy (RT). Methods: We designed and conducted a two-phase study for two disease sites and two treatment modalities-adaptive RT for non-small cell lung cancer (NSCLC) and adaptive stereotactic body RT for hepatocellular carcinoma (HCC)-in which clinicians were asked to consider mid-treatment modification of the dose per fraction for a number of retrospective cancer patients without AI-support (Unassisted Phase) and with AI-assistance (AI-assisted Phase). The AI-CDSS graphically presented trade-offs in tumor control and the likelihood of toxicity to organs at risk, provided an optimal recommendation, and associated model uncertainties. In addition, we asked for clinicians' decision confidence level and trust level in individual AI recommendations and encouraged them to provide written remarks. We enrolled 13 evaluators (radiation oncology physicians and residents) from two medical institutions located in two different states, out of which, 4 evaluators volunteered in both NSCLC and HCC studies, resulting in a total of 17 completed evaluations (9 NSCLC, and 8 HCC). To limit the evaluation time to under an hour, we selected 8 treated patients for NSCLC and 9 for HCC, resulting in a total of 144 sets of evaluations (72 from NSCLC and 72 from HCC). Evaluation for each patient consisted of 8 required inputs and 2 optional remarks, resulting in up to a total of 1440 data points. Results: AI-assistance did not homogeneously influence all experts and clinical decisions. From NSCLC cohort, 41 (57%) decisions and from HCC cohort, 34 (47%) decisions were adjusted after AI assistance. Two evaluations (12%) from the NSCLC cohort had zero decision adjustments, while the remaining 15 (88%) evaluations resulted in at least two decision adjustments. Decision adjustment level positively correlated with dissimilarity in decision-making with AI [NSCLC: ρ = 0.53 ( p < 0.001); HCC: ρ = 0.60 ( p < 0.001)] indicating that evaluators adjusted their decision closer towards AI recommendation. Agreement with AI-recommendation positively correlated with AI Trust Level [NSCLC: ρ = 0.59 ( p < 0.001); HCC: ρ = 0.7 ( p < 0.001)] indicating that evaluators followed AI's recommendation if they agreed with that recommendation. The correlation between decision confidence changes and decision adjustment level showed an opposite trend [NSCLC: ρ = -0.24 ( p = 0.045), HCC: ρ = 0.28 ( p = 0.017)] reflecting the difference in behavior due to underlying differences in disease type and treatment modality. Decision confidence positively correlated with the closeness of decisions to the standard of care (NSCLC: 2 Gy/fx; HCC: 10 Gy/fx) indicating that evaluators were generally more confident in prescribing dose fractionations more similar to those used in standard clinical practice. Inter-evaluator agreement increased with AI-assistance indicating that AI-assistance can decrease inter-physician variability. The majority of decisions were adjusted to achieve higher tumor control in NSCLC and lower normal tissue complications in HCC. Analysis of evaluators' remarks indicated concerns for organs at risk and RT outcome estimates as important decision-making factors. Conclusions: Human-AI interaction depends on the complex interrelationship between expert's prior knowledge and preferences, patient's state, disease site, treatment modality, model transparency, and AI's learned behavior and biases. The collaborative decision-making process can be summarized as follows: (i) some clinicians may not believe in an AI system, completely disregarding its recommendation, (ii) some clinicians may believe in the AI system but will critically analyze its recommendations on a case-by-case basis; (iii) when a clinician finds that the AI recommendation indicates the possibility for better outcomes they will adjust their decisions accordingly; and (iv) When a clinician finds that the AI recommendation indicate a worse possible outcome they will disregard it and seek their own alternative approach.

Prospective Evaluation of Limited-Stage Small Cell Lung Cancer Radiotherapy Fractionation Regimen Usage and Acute Toxicity in a Large Statewide Quality Collaborative.

PURPOSE: National guidelines on limited-stage small cell lung cancer (LS-SCLC) treatment give preference to a hyperfractionated regimen of 45 Gy in 30 fractions delivered twice daily; however, use of this regimen is uncommon compared with once-daily regimens. The purpose of this study was to characterize the LS-SCLC fractionation regimens used throughout a statewide collaborative, analyze patient and treatment factors associated with these regimens, and describe real-world acute toxicity profiles of once- and twice-daily radiation therapy (RT) regimens. METHODS AND MATERIALS: Demographic, clinical, and treatment data along with physician-assessed toxicity and patient-reported outcomes were prospectively collected by 29 institutions within the Michigan Radiation Oncology Quality Consortium between 2012 and 2021 for patients with LS-SCLC. We modeled the influence of RT fractionation and other patient-level variables clustered by treatment site on the odds of a treatment break specifically due to toxicity with multilevel logistic regression. National Cancer Institute Common Terminology Criteria for Adverse Events, version 4.0, incident grade 2 or worse toxicity was longitudinally compared between regimens. RESULTS: There were 78 patients (15.6% overall) treated with twice-daily RT and 421 patients treated with once-daily RT. Patients receiving twice-daily RT were more likely to be married or living with someone (65% vs 51%; P = .019) and to have no major comorbidities (24% vs 10%; P = .017). Once-daily RT fractionation toxicity peaked during RT, and twice-daily toxicity peaked within 1 month after RT. After stratifying by treatment site and adjusting for patient-level variables, once-daily treated patients had 4.11 (95% confidence interval, 1.31-12.87) higher odds of treatment break specifically due to toxicity than twice-daily treated patients. CONCLUSIONS: Hyperfractionation for LS-SCLC remains infrequently prescribed despite the lack of evidence demonstrating superior efficacy or lower toxicity of once-daily RT. With peak acute toxicity after RT and lower likelihood of a treatment break with twice-daily fractionation in real-word practice, providers may start using hyperfractionated RT more frequently.

A clinical decision support system for AI-assisted decision-making in response-adaptive radiotherapy (ARCliDS).

Involvement of many variables, uncertainty in treatment response, and inter-patient heterogeneity challenge objective decision-making in dynamic treatment regime (DTR) in oncology. Advanced machine learning analytics in conjunction with information-rich dense multi-omics data have the ability to overcome such challenges. We have developed a comprehensive artificial intelligence (AI)-based optimal decision-making framework for assisting oncologists in DTR. In this work, we demonstrate the proposed framework to Knowledge Based Response-Adaptive Radiotherapy (KBR-ART) applications by developing an interactive software tool entitled Adaptive Radiotherapy Clinical Decision Support (ARCliDS). ARCliDS is composed of two main components: Artifcial RT Environment (ARTE) and Optimal Decision Maker (ODM). ARTE is designed as a Markov decision process and modeled via supervised learning. Given a patient's pre- and during-treatment information, ARTE can estimate treatment outcomes for a selected daily dosage value (radiation fraction size). ODM is formulated using reinforcement learning and is trained on ARTE. ODM can recommend optimal daily dosage adjustments to maximize the tumor local control probability and minimize the side effects. Graph Neural Networks (GNN) are applied to exploit the inter-feature relationships for improved modeling performance and a novel double GNN architecture is designed to avoid nonphysical treatment response. Datasets of size 117 and 292 were available from two clinical trials on adaptive RT in non-small cell lung cancer (NSCLC) patients and adaptive stereotactic body RT (SBRT) in hepatocellular carcinoma (HCC) patients, respectively. For training and validation, dense data with 297 features were available for 67 NSCLC patients and 110 features for 71 HCC patients. To increase the sample size for ODM training, we applied Generative Adversarial Networks to generate 10,000 synthetic patients. The ODM was trained on the synthetic patients and validated on the original dataset. We found that, Double GNN architecture was able to correct the nonphysical dose-response trend and improve ARCliDS recommendation. The average root mean squared difference (RMSD) between ARCliDS recommendation and reported clinical decisions using double GNNs were 0.61 [0.03] Gy/frac (mean [sem]) for adaptive RT in NSCLC patients and 2.96 [0.42] Gy/frac for adaptive SBRT HCC compared to the single GNN's RMSDs of 0.97 [0.12] Gy/frac and 4.75 [0.16] Gy/frac, respectively. Overall, For NSCLC and HCC, ARCliDS with double GNNs was able to reproduce 36% and 50% of the good clinical decisions (local control and no side effects) and improve 74% and 30% of the bad clinical decisions, respectively. In conclusion, ARCliDS is the first web-based software dedicated to assist KBR-ART with multi-omics data. ARCliDS can learn from the reported clinical decisions and facilitate AI-assisted clinical decision-making for improving the outcomes in DTR.

Real-time, volumetric imaging of radiation dose delivery deep into the liver during cancer treatment.

Ionizing radiation acoustic imaging (iRAI) allows online monitoring of radiation's interactions with tissues during radiation therapy, providing real-time, adaptive feedback for cancer treatments. We describe an iRAI volumetric imaging system that enables mapping of the three-dimensional (3D) radiation dose distribution in a complex clinical radiotherapy treatment. The method relies on a two-dimensional matrix array transducer and a matching multi-channel preamplifier board. The feasibility of imaging temporal 3D dose accumulation was first validated in a tissue-mimicking phantom. Next, semiquantitative iRAI relative dose measurements were verified in vivo in a rabbit model. Finally, real-time visualization of the 3D radiation dose delivered to a patient with liver metastases was accomplished with a clinical linear accelerator. These studies demonstrate the potential of iRAI to monitor and quantify the 3D radiation dose deposition during treatment, potentially improving radiotherapy treatment efficacy using real-time adaptive treatment.

Gender-Based Discrimination and Sexual Harassment in Medical Physics.

PURPOSE: Gender-based discrimination and sexual harassment have been well-studied in the fields of science, technology, engineering, math, and medicine. However, less is known about these topics and their effect within the profession of medical physics. We aimed to better understand and clarify the views and experiences of practicing medical physicists and medical physics residents regarding gender-based discrimination and sexual harassment. METHODS AND MATERIALS: We conducted in-depth, semistructured, and confidential interviews with 32 practicing medical physicists and medical physics residents across the United States. The interviews were broad and covered the topics of discrimination, mentorship, and work/life integration. All participants were associated with a department with a residency program accredited by the Commission on Accreditation of Medical Physics Education Programs and had appointments with a clinical component. RESULTS: Participants shared views about gender-based discrimination and sexual harassment that were polarized. Some perceived that discrimination and harassment were a current concern within medical physics, while some either perceived that they were not a concern or that discrimination positively affected women and minoritized populations. Many participants shared personal experiences of discrimination and harassment, including those related to unequal compensation, discrimination against mothers, discrimination during the hiring process, gender-biased assumptions about behaviors or goals, communication biases, and overt and persistent sexual harassment. CONCLUSIONS: There is an urgent need to acknowledge, better understand, and address gender-based discrimination and sexual harassment in the field of medical physics.

NRG Oncology/RTOG1205: A Randomized Phase II Trial of Concurrent Bevacizumab and Reirradiation Versus Bevacizumab Alone as Treatment for Recurrent Glioblastoma.

PURPOSE: To assess whether reirradiation (re-RT) and concurrent bevacizumab (BEV) improve overall survival (OS) and/or progression-free survival (PFS), compared with BEV alone in recurrent glioblastoma (GBM). The primary objective was OS, and secondary objectives included PFS, response rate, and treatment adverse events (AEs) including delayed CNS toxicities. METHODS: NRG Oncology/RTOG1205 is a prospective, phase II, randomized trial of re-RT and BEV versus BEV alone. Stratification factors included age, resection, and Karnofsky performance status (KPS). Patients with recurrent GBM with imaging evidence of tumor progression ≥ 6 months from completion of prior chemo-RT were eligible. Patients were randomly assigned 1:1 to re-RT, 35 Gy in 10 fractions, with concurrent BEV IV 10 mg/kg once in every 2 weeks or BEV alone until progression. RESULTS: From December 2012 to April 2016, 182 patients were randomly assigned, of whom 170 were eligible. Patient characteristics were well balanced between arms. The median follow-up for censored patients was 12.8 months. There was no improvement in OS for BEV + RT, hazard ratio, 0.98; 80% CI, 0.79 to 1.23; P = .46; the median survival time was 10.1 versus 9.7 months for BEV + RT versus BEV alone. The median PFS for BEV + RT was 7.1 versus 3.8 months for BEV, hazard ratio, 0.73; 95% CI, 0.53 to 1.0; P = .05. The 6-month PFS rate improved from 29.1% (95% CI, 19.1 to 39.1) for BEV to 54.3% (95% CI, 43.5 to 65.1) for BEV + RT, P = .001. Treatment was well tolerated. There were a 5% rate of acute grade 3+ treatment-related AEs and no delayed high-grade AEs. Most patients died of recurrent GBM. CONCLUSION: To our knowledge, NRG Oncology/RTOG1205 is the first prospective, randomized multi-institutional study to evaluate the safety and efficacy of re-RT in recurrent GBM using modern RT techniques. Overall, re-RT was shown to be safe and well tolerated. BEV + RT demonstrated a clinically meaningful improvement in PFS, specifically the 6-month PFS rate but no difference in OS.

Development and Clinical Implementation of an Automated Virtual Integrative Planner for Radiation Therapy of Head and Neck Cancer.

Purpose: Head and neck (HN) radiation (RT) treatment planning is complex and resource intensive. Deviations and inconsistent plan quality significantly affect clinical outcomes. We sought to develop a novel automated virtual integrative (AVI) knowledge-based planning application to reduce planning time, increase consistency, and improve baseline quality. Methods and Materials: An in-house write-enabled script was developed from a library of 668 previously treated HN RT plans. Prospective hazard analysis was performed, and mitigation strategies were implemented before clinical release. The AVI-planner software was retrospectively validated in a cohort of 52 recent HN cases. A physician panel evaluated planning limitations during initial deployment, and feedback was enacted via software refinements. A final second set of plans was generated and evaluated. Kolmogorov-Smirnov test in addition to generalized evaluation metric and weighted experience score were used to compare normal tissue sparing between final AVI planner versus respective clinically treated and historically accepted plans. A t test was used to compare the interactive time, complexity, and monitor units for AVI planner versus manual optimization. Results: Initially, 86% of plans were acceptable to treat, with 10% minor and 4% major revisions or rejection recommended. Variability was noted in plan quality among HN subsites, with high initial quality for oropharynx and oral cavity plans. Plans needing revisions were comprised of sinonasal, nasopharynx, P-16 negative squamous cell carcinoma unknown primary, or cutaneous primary sites. Normal tissue sparing varied within subsites, but AVI planner significantly lowered mean larynx dose (median, 18.5 vs 19.7 Gy; P < .01) compared with clinical plans. AVI planner significantly reduced interactive optimization time (mean, 2 vs 85 minutes; P < .01). Conclusions: AVI planner reliably generated clinically acceptable RT plans for oral cavity, salivary, oropharynx, larynx, and hypopharynx cancers. Physician-driven iterative learning processes resulted in favorable evolution in HN RT plan quality with significant time savings and improved consistency using AVI planner.

A human-in-the-loop based Bayesian network approach to improve imbalanced radiation outcomes prediction for hepatocellular cancer patients with stereotactic body radiotherapy.

Background: Imbalanced outcome is one of common characteristics of oncology datasets. Current machine learning approaches have limitation in learning from such datasets. Here, we propose to resolve this problem by utilizing a human-in-the-loop (HITL) approach, which we hypothesize will also lead to more accurate and explainable outcome prediction models. Methods: A total of 119 HCC patients with 163 tumors were used in the study. 81 patients with 104 tumors from the University of Michigan Hospital treated with SBRT were considered as a discovery dataset for radiation outcomes model building. The external testing dataset included 59 tumors from 38 patients with SBRT from Princess Margaret Hospital. In the discovery dataset, 100 tumors from 77 patients had local control (LC) (96% of 104 tumors) and 23 patients had at least one grade increment of ALBI (I-ALBI) during six-month follow up (28% of 81 patients). Each patient had a total of 110 features, where 15 or 20 features were identified by physicians as expert knowledge features (EKFs) for LC or I-ALBI prediction. We proposed a HITL based Bayesian network (HITL-BN) approach to enhance the capability of selecting important features from imbalanced data in terms of accuracy and explainability through humans' participation by integrating feature importance ranking and Markov blanket algorithms. A pure data-driven Bayesian network (PD-BN) method was applied to the same discovery dataset of HCC patients as a benchmark. Results: In the training and testing phases, the areas under receiver operating characteristic curves of the HITL-BN models for LC or I-ALBI prediction during SBRT are 0.85 (95% confidence interval: 0.75-0.95) or 0.89 (0.81-0.95) and 0.77 or 0.78, respectively. They significantly outperformed the during-treatment PD-BN model in predicting LC or I-ALBI based on the discovery cross-validation and testing datasets from the Delong tests. Conclusion: By allowing the human expert to be part of the model building process, the HITL-BN approach yielded significantly improved accuracy as well as better explainability when dealing with imbalanced outcomes in the prediction of post-SBRT treatment response of HCC patients when compared to the PD-BN method.