ABSTRACT
BACKGROUND: The reliable and efficient estimation of uncertainty in artificial intelligence (AI) models poses an ongoing challenge in many fields such as radiation therapy. AI models are intended to automate manual steps involved in the treatment planning workflow. We focus in this study on dose prediction models that predict an optimal dose trade-off for each new patient for a specific treatment modality. They can guide physicians in the optimization, be part of automatic treatment plan generation or support decision in treatment indication. Most common uncertainty estimation methods are based on Bayesian approximations, like Monte Carlo dropout (MCDO) or Deep ensembling (DE). These two techniques, however, have a high inference time (i.e., require multiple inference passes) and might not work for detecting out-of-distribution (OOD) data (i.e., overlapping uncertainty estimate for in-distribution (ID) and OOD). PURPOSE: In this study, we present a direct uncertainty estimation method and apply it for a dose prediction U-Net architecture. It can be used to flag OOD data and give information on the quality of the dose prediction. METHODS: Our method consists in the addition of a branch decoding from the bottleneck which reconstructs the CT scan given as input. The input reconstruction error can be used as a surrogate of the model uncertainty. For the proof-of-concept, our method is applied to proton therapy dose prediction in head and neck cancer patients. A dataset of 60 oropharyngeal patients was used to train the network using a nested cross-validation approach with 11 folds (training: 50 patients, validation: 5 patients, test: 5 patients). For the OOD experiment, we used 10 extra patients with a different head and neck sub-location. Accuracy, time-gain, and OOD detection are analyzed for our method in this particular application and compared with the popular MCDO and DE. RESULTS: The additional branch did not reduce the accuracy of the dose prediction model. The median absolute error is close to zero for the target volumes and less than 1% of the dose prescription for organs at risk. Our input reconstruction method showed a higher Pearson correlation coefficient with the prediction error (0.620) than DE (0.447) and MCDO (between 0.599 and 0.612). Moreover, our method allows an easier identification of OOD (no overlap for ID and OOD data and a Z-score of 34.05). The uncertainty is estimated simultaneously to the regression task, therefore requires less time and computational resources. CONCLUSIONS: This study shows that the error in the CT scan reconstruction can be used as a surrogate of the uncertainty of the model. The Pearson correlation coefficient with the dose prediction error is slightly higher than state-of-the-art techniques. OOD data can be more easily detected and the uncertainty metric is computed simultaneously to the regression task, therefore faster than MCDO or DE. The code and pretrained model are available on the gitlab repository: https://gitlab.com/ai4miro/ct-reconstruction-for-uncertainty-quatification-of-hdunet.
ABSTRACT
RATIONALE & OBJECTIVE: Continuous kidney replacement therapy (CKRT) is preferred when available for hemodynamically unstable acute kidney injury (AKI) patients in the intensive care unit (ICU). The Kidney Disease: Improving Global Outcomes (KDIGO) guidelines recommend a delivered CKRT dose of 20-25mL/kg/h; however, in Japan the doses are typically below this recommendation due to government health insurance system restrictions. This study investigated the association between mortality and dose of CKRT. STUDY DESIGN: Single-center retrospective cohort study. SETTING & PARTICIPANTS: Critically ill patients with AKI treated with CKRT at a tertiary Japanese university hospital between January 1, 2012, and December 31, 2021. EXPOSURE: Delivered CKRT doses below or above the median. OUTCOME: 90-day mortality after CKRT initiation. ANALYTICAL APPROACH: Multivariable Cox regression analysis and Kaplan-Meier analysis. RESULTS: The study population consisted of 494 patients. The median age was 72 years, and 309 patients (62.6%) were men. Acute tubular injury was the leading cause of AKI, accounting for 81.8%. The median delivered CKRT dose was 13.2mL/kg/h. Among the study participants, 456 (92.3%) received delivered CKRT doses below 20mL/kg/h, and 204 (41.3%) died within 90 days after CKRT initiation. Multivariable Cox regression analysis revealed increased mortality in the below-median group (HR, 1.73 [95% CI, 1.19-2.51], P=0.004). Additionally, a significant, inverse, nonlinear association between 90-day mortality and delivered CKRT dose was observed using delivered CKRT dose as a continuous variable. LIMITATIONS: Single-center, retrospective, observational study. CONCLUSIONS: A lower delivered CKRT dose was independently associated with higher 90-day mortality among critically ill patients who mostly received dosing below the current KDIGO recommendations. PLAIN-LANGUAGE SUMMARY: The Kidney Disease: Improving Global Outcomes (KDIGO) guidelines recommend delivering a continuous kidney replacement therapy (CKRT) dose of 20-25mL/kg/h. However, it is not clear if it is safe to use delivered CKRT doses below this recommendation. In this study, over 90% of the patients received CKRT with a delivered dose below the KDIGO recommendation. We divided these patients into 2 groups based on the median delivered CKRT dose. Our findings show that a delivered CKRT dose below the median was associated with increased risk of death within 90 days. These findings show that a lower delivered CKRT dose was independently associated with higher 90-day mortality among critically ill patients who mostly received dosing below current KDIGO recommendations.
Subject(s)
Acute Kidney Injury , Continuous Renal Replacement Therapy , Critical Illness , Humans , Male , Acute Kidney Injury/mortality , Acute Kidney Injury/therapy , Female , Retrospective Studies , Critical Illness/mortality , Critical Illness/therapy , Aged , Continuous Renal Replacement Therapy/methods , Middle Aged , Cohort Studies , Japan/epidemiology , Intensive Care Units , Aged, 80 and overABSTRACT
BACKGROUND: Variable relative biological effectiveness (RBE) models in treatment planning have been proposed to optimize the therapeutic ratio of proton therapy. It has been reported that proton RBE decreases with increasing tumor oxygen level, offering an opportunity to address hypoxia-related radioresistance with RBE-weighted optimization. PURPOSE: Here, we obtain a voxel-level estimation of partial oxygen pressure to weigh RBE values in a single biologically informed beam orientation optimization (BOO) algorithm. METHODS: Three glioblastoma patients with [18 F]-fluoromisonidazole (FMISO)-PET/CT images were selected from the institutional database. Oxygen values were derived from tracer uptake using a nonlinear least squares curve fitting. McNamara RBE, calculated from proton dose, was then weighed using oxygen enhancement ratios (OER) for each voxel and incorporated into the dose fidelity term of the BOO algorithm. The nonlinear optimization problem was solved using a split-Bregman approach, with FISTA as the solver. The proposed hypoxia informed RBE-weighted method (HypRBE) was compared to dose fidelity terms using the constant RBE of 1.1 (cRBE) and the normoxic McNamara RBE model (RegRBE). Tumor homogeneity index (HI), maximum biological dose (Dmax), and D95%, as well as OAR therapeutic index (TI = gEUDCTV /gEUDOAR ) were evaluated along with worst-case statistics after normalization to normal tissue isotoxicity. RESULTS: Compared to [cRBE, RegRBE], HypRBE increased tumor HI, Dmax, and D95% across all plans by on average [31.3%, 31.8%], [48.6%, 27.1%], and [50.4%, 23.8%], respectively. In the worst-case scenario, the parameters increase on average by [12.5%, 14.7%], [7.3%,-8.9%], and [22.3%, 2.1%]. Despite increased OAR Dmean and Dmax by [8.0%, 3.0%] and [13.1%, -0.1%], HypRBE increased average TI by [22.0%, 21.1%]. Worst-case OAR Dmean, Dmax, and TI worsened by [17.9%, 4.3%], [24.5%, -1.2%], and [9.6%, 10.5%], but in the best cases, HypRBE escalates tumor coverage significantly without compromising OAR dose, increasing the therapeutic ratio. CONCLUSIONS: We have developed an optimization algorithm whose dose fidelity term accounts for hypoxia-informed RBE values. We have shown that HypRBE selects bE:\Alok\aaeams better suited to deliver high physical dose to low RBE, hypoxic tumor regions while sparing the radiosensitive normal tissue.
Subject(s)
Glioblastoma , Proton Therapy , Humans , Proton Therapy/methods , Protons , Relative Biological Effectiveness , Positron Emission Tomography Computed Tomography , Radiotherapy Planning, Computer-Assisted/methods , Hypoxia/radiotherapy , Oxygen , Radiotherapy DosageABSTRACT
Significance: Photodynamic therapy (PDT) is an established cancer treatment utilizing light-activated photosensitizers (PS). Effective treatment hinges on the PDT dose-dependent on PS concentration and light fluence-delivered over time. We introduce an innovative eight-channel PDT dose dosimetry system capable of concurrently measuring light fluence and PS concentration during treatment. Aim: We aim to develop and evaluate an eight-channel PDT dose dosimetry system for simultaneous measurement of light fluence and PS concentration. By addressing uncertainties due to tissue variations, the system enhances accurate PDT dosimetry for improved treatment outcomes. Approach: The study positions eight isotropic detectors strategically within the pleural cavity before PDT. These detectors are linked to bifurcated fibers, distributing signals to both a photodiode and a spectrometer. Calibration techniques are applied to counter tissue-related variations and improve measurement accuracy. The fluorescence signal is normalized using the measured light fluence, compensating for variations in tissue properties. Measurements were taken in 78 sites in the pleural cavities of 20 patients. Results: Observations reveal minimal Photofrin concentration variation during PDT at each site, juxtaposed with significant intra- and inter-patient heterogeneities. Across 78 treated sites in 20 patients, the average Photofrin concentration for all 78 sites is 4.98 µM, with a median concentration of 4.47 µM. The average PDT dose for all 78 sites is 493.17 µMJ/cm2, with a median dose of 442.79 µMJ/cm2. A significant variation in PDT doses is observed, with a maximum difference of 3.1 times among all sites within one patient and a maximum difference of 9.8 times across all patients. Conclusions: The introduced eight-channel PDT dose dosimetry system serves as a valuable real-time monitoring tool for light fluence and PS concentration during PDT. Its ability to mitigate uncertainties arising from tissue properties enhances dosimetry accuracy, thus optimizing treatment outcomes and bolstering the effectiveness of PDT in cancer therapy.
Subject(s)
Dihematoporphyrin Ether , Photochemotherapy , Humans , Dihematoporphyrin Ether/therapeutic use , Photochemotherapy/methods , Photosensitizing Agents/therapeutic use , Radiometry/methodsABSTRACT
INTRODUCTION: There is limited evidence on the costs and outcomes of patients with aphasia after stroke. The aim of this study was to estimate costs in patients with aphasia after stroke according to the aphasia therapies provided. METHODS: A three-arm, prospective, randomized, parallel group, open-label, blinded endpoint assessment trial conducted in Australia and New Zealand. Usual ward-based care (Usual Care) was compared to additional usual ward-based therapy (Usual Care Plus) and a prescribed and structured aphasia therapy program in addition to Usual Care (the VERSE intervention). Information about healthcare utilization and productivity were collected to estimate costs in Australian dollars for 2017-18. Multivariable regression models with bootstrapping were used to estimate differences in costs and outcomes (clinically meaningful change in aphasia severity measured by the WAB-R-AQ). RESULTS: Overall, 202/246 (82%) participants completed follow-up at 26 weeks. Median costs per person were $23,322 (Q1 5,367, Q3 52,669, n = 63) for Usual Care, $26,923 (Q1 7,303, Q3 76,174, n = 70) for Usual Care Plus and $31,143 (Q1 7,001. Q3 62,390, n = 69) for VERSE. No differences in costs and outcomes were detected between groups. Usual Care Plus was inferior (i.e. more costly and less effective) in 64% of iterations, and in 18% was less costly and less effective compared to Usual Care. VERSE was inferior in 65% of samples and less costly and less effective in 12% compared to Usual Care. CONCLUSION: There was limited evidence that additional intensively delivered aphasia therapy within the context of usual acute care provided was worthwhile in terms of costs for the outcomes gained.
Subject(s)
Aphasia , Stroke Rehabilitation , Stroke , Humans , Stroke/complications , Stroke/therapy , Cost-Benefit Analysis , Prospective Studies , Speech , Australia , Aphasia/etiology , Aphasia/rehabilitationABSTRACT
PURPOSE: Empirical relative biological effectiveness (RBE) models have been used to estimate the biological dose in proton therapy but do not adequately capture the factors influencing RBE values for treatment planning. We reformulate the McNamara RBE model such that it can be added as a linear biological dose fidelity term within our previously developed sensitivity-regularized and heterogeneity-weighted beam orientation optimization (SHBOO) framework. METHODS: Based on our SHBOO framework, we formulated the biological optimization problem to minimize total McNamara RBE dose to OARs. We solve this problem using two optimization algorithms: FISTA (McNam-FISTA) and Chambolle-Pock (McNam-CP). We compare their performances with a physical dose optimizer assuming RBE = 1.1 in all structures (PHYS-FISTA) and an LET-weighted dose model (LET-FISTA). Three head and neck patients were planned with the four techniques and compared on dosimetry and robustness. RESULTS: Compared to Phys-FISTA, McNam-CP was able to match CTV [HI, Dmax, D95%, D98%] by [0.00, 0.05%, 1.4%, 0.8%]. McNam-FISTA and McNam-CP were able to significantly improve overall OAR [Dmean, Dmax] by an average of [36.1%,26.4%] and [29.6%, 20.3%], respectively. Regarding CTV robustness, worst [Dmax, V95%, D95%, D98%] improvement of [-6.6%, 6.2%, 6.0%, 4.8%] was reported for McNam-FISTA and [2.7%, 2.7%, 5.3%, -4.3%] for McNam-CP under combinations of range and setup uncertainties. For OARs, worst [Dmax, Dmean] were improved by McNam-FISTA and McNam-CP by an average of [25.0%, 19.2%] and [29.5%, 36.5%], respectively. McNam-FISTA considerably improved dosimetry and CTV robustness compared to LET-FISTA, which achieved better worst-case OAR doses. CONCLUSION: The four optimization techniques deliver comparable biological doses for the head and neck cases. Besides modest CTV coverage and robustness improvement, OAR biological dose and robustness were substantially improved with both McNam-FISTA and McNam-CP, showing potential benefit for directly incorporating McNamara RBE in proton treatment planning.
Subject(s)
Proton Therapy , Radiotherapy, Intensity-Modulated , Humans , Organs at Risk , Proton Therapy/methods , Radiotherapy Dosage , Radiotherapy Planning, Computer-Assisted/methods , Radiotherapy, Intensity-Modulated/methods , Relative Biological EffectivenessABSTRACT
@#Introduction: Increase in the number of opioids seized in the recent year may indicate increased opioid use in Malaysia. In counteracting opioid abuse, Methadone Maintenance Therapy (MMT) was introduced in Malaysia but relapse following MMT has become an important issue. This study aimed to determine the prevalence and patient factors that served as predictors of opioid relapse among MMT patients. Method: A cross-sectional study involving 159 MMT patients who have reached dose stabilization (eight weeks at a constant dose of methadone) was conducted in Johor Bahru Health District. The dependent variable was opioid relapse, while the independent variables include socio-demographic characteristics, MMT history, crime history, cognitive and interpersonal factors, and social-environment influence. Face-to-face interviews using structured questionnaires and secondary data collection using data collection sheets were done. Multiple logistic regression was used to determine the predictors. Significant level set at alpha less than 5%. Result: The response rate was 86.9% with majority of them were Malay, male, and Muslim. The prevalence of opioid relapse was 11.9%. Those who were non-polydrug users (AOR=3.701, 95%CI=1.182, 11.587, p=0.025), classified as having moderate (AOR=5.869, 95%CI=1.524, 22.595, p=0.010) and high (AOR=5.952, 95%CI=1.000, 35.445, p=0.050) relapse risk response after given hypothetical situation whether respondent been offered drug or not, were more likely to have relapsed. Respondents with higher cognitive and behavioral problem-solving response scores were less likely to have relapsed (AOR=0.949, 95%CI=0.909, 0.991, p=0.008). Conclusion: About 1 in 5 MMT clients had relapsed after they reach dose stabilization. The predictors of opioid relapse were non-polydrug users, having moderate to high relapse risk, and cognitive and behavioral problem-solving responses.
ABSTRACT
PURPOSE: Radiotherapy, especially with charged particles, is sensitive to executional and preparational uncertainties that propagate to uncertainty in dose and plan quality indicators, for example, dose-volume histograms (DVHs). Current approaches to quantify and mitigate such uncertainties rely on explicitly computed error scenarios and are thus subject to statistical uncertainty and limitations regarding the underlying uncertainty model. Here we present an alternative, analytical method to approximate moments, in particular expectation value and (co)variance, of the probability distribution of DVH-points, and evaluate its accuracy on patient data. METHODS: We use Analytical Probabilistic Modeling (APM) to derive moments of the probability distribution over individual DVH-points based on the probability distribution over dose. By using the computed moments to parameterize distinct probability distributions over DVH-points (here normal or beta distributions), not only the moments but also percentiles, that is, α - DVHs, are computed. The model is subsequently evaluated on three patient cases (intracranial, paraspinal, prostate) in 30- and single-fraction scenarios by assuming the dose to follow a multivariate normal distribution, whose moments are computed in closed-form with APM. The results are compared to a benchmark based on discrete random sampling. RESULTS: The evaluation of the new probabilistic model on the three patient cases against a sampling benchmark proves its correctness under perfect assumptions as well as good agreement in realistic conditions. More precisely, ca. 90% of all computed expected DVH-points and their standard deviations agree within 1% volume with their empirical counterpart from sampling computations, for both fractionated and single fraction treatments. When computing α - DVH, the assumption of a beta distribution achieved better agreement with empirical percentiles than the assumption of a normal distribution: While in both cases probabilities locally showed large deviations (up to ±0.2), the respective - DVHs for α={0.05,0.5,0.95} only showed small deviations in respective volume (up to ±5% volume for a normal distribution, and up to 2% for a beta distribution). A previously published model from literature, which was included for comparison, exhibited substantially larger deviations. CONCLUSIONS: With APM we could derive a mathematically exact description of moments of probability distributions over DVH-points given a probability distribution over dose. The model generalizes previous attempts and performs well for both choices of probability distributions, that is, normal or beta distributions, over DVH-points.
Subject(s)
Models, Statistical , Radiotherapy Planning, Computer-Assisted , Humans , Male , Normal Distribution , Probability , Radiotherapy DosageABSTRACT
Selecting the dose of a targeted cancer medicine that is most appropriate for a specific individual is a rational approach to maximize therapeutic outcomes and minimize toxicity. There are many different options for optimizing the dose of targeted cancer medicines and the purpose of this review is to provide a comprehensive comparison of the main options explored in prospective studies. Precision initial dose selection of targeted cancer therapies has been minimally explored to date; however, concentration, toxicity, and therapeutic outcome markers are used to guide on-therapy dose adaption of targeted cancer therapies across several medicines and cancers. While a specific concentration, toxicity, or therapeutic outcome marker commonly dominates an investigated precision on-therapy dose adaption strategy, greater attention to simultaneously account for exposure, toxicity, therapeutic outcomes, disease status, time since treatment initiation and patient preferences are required for optimal patient outcomes. To enable successful implementation of precision dosing strategies for targeted cancer medicines into clinical practice, future prospective studies aiming to develop strategies should consider these elements in their design.
Subject(s)
Antineoplastic Agents/administration & dosage , Molecular Targeted Therapy , Neoplasms/drug therapy , Antineoplastic Agents/adverse effects , Dose-Response Relationship, Drug , Humans , Precision Medicine , Treatment OutcomeABSTRACT
PURPOSE: The objective was to characterize both systematic and random errors in Positron Emission Tomography (PET)-based 90 Y three-dimensional (3D) dose quantification. METHODS: A modified NEMA-IEC phantom was used to emulate 90 Y-microsphere PET imaging conditions: sphere activity concentrations of 1.6 and 4.8 MBq/cc, sphere-to-background ratios of 4 and 13, and sphere diameters of 13, 17, and 37 mm. PET data were acquired using a GE D690 PET/CT scanner for 300 min on days 0-11. The data were downsampled to 60-5 min for multiple realizations to evaluate the count starvation effect. The image reconstruction algorithm was 3D-OSEM with PSF + TOF modeling; the parameters were optimized for dose-volume histogram (DVH), as a 90 Y 3D dose quantification. 90 Y-PET images were converted to dose maps using the local deposition method, then the sphere DVHs were calculated. The ground truth for the DVH was calculated using convolution method. Dose linearity was evaluated in decaying 90 Y activity (reduced count rate and total count) and decreasing acquisition durations (reduced total count only). Finally, the impacts of the low 32-ppm positron yield on PET-based 3D 90 Y-dose quantification were evaluated; the bias and variability of resulting DVHs were characterized. RESULTS: We observed nonlinear errors that depended on the 90 Y activity (count rate) and not on the total true prompt counts. These nonlinear errors in mean dose underestimated the measured mean dose by> 20% for a measured dose range of 40-230 Gy; although the shapes of the DVH were not altered. Compensation based on empirical models reduced the nonlinearity errors to be within 5% for measured dose range of 40-230 Gy. In contrast, the errors due to nonuniformity introduced by image noise dominated the systematic errors in the DVH and stretched the DVH on both tails. For the 37-mm sphere, the magnitude of errors in D80 increased from -25% to -36% when acquisition duration was decreased from 300 to 10 min. The effect of image noise on DVH was more extensive in smaller spheres; for the 17-mm sphere, the magnitude of errors in D80 increased from -29% to -45% acquisition duration was decreased from 300 to 10 min. For the 37-mm sphere, the errors in D20 increased from +3.5% to only +10.5% when the acquisition duration was decreased from 300 to 10 min; in the 17-mm sphere, the errors in D20 were 6.5% for both 300- and 10-min sphere images. CONCLUSIONS: Count-starved 90 Y-PET data introduce both systematic and random errors. The systematic error increases the apparent nonuniformity of the DVH, while the random error increases the uncertainty in the DVH. The systematic errors were larger than the random errors. Lower count rate of 90 Y-PET also introduces systematic bias, which is scanner specific. The errors of bias-compensated mean tumor dose were <10% when 90 Y-PET scan time was >15 min/bed for tumors >37 mm. Dmedian and Dmean were the most stable dose metrics. An acquisition duration of 30 min is recommended to keep the random errors < 10% for a typical tumor with sphere equivalent diameter >17 mm and average tumor dose >40 Gy.
Subject(s)
Positron Emission Tomography Computed Tomography , Positron-Emission Tomography , Algorithms , Image Processing, Computer-Assisted , Phantoms, ImagingABSTRACT
INTRODUCTION: Radiation Therapy Oncology Group (RTOG) report #0813 and 0915 recommends using D2cm and R50% as plan quality metrics for evaluation of normal tissue sparing in stereotactic body radiation therapy (SBRT) of lung lesion. This study introduces dose falloff gradient (DFG) as a tool for analyzing the dose beyond the planning target volume (PTV) extending into normal tissue structures. In ascertaining the impact of PTV size and SBRT planning techniques in DFG, this study questions the independence of the RTOG recommended metrics. MATERIALS AND METHODS: In this retrospective study, 41 RapidArc lung SBRT plans with 2 or 3 complete or partial arcs were analyzed. PTV volumes ranged between 5.3 and 113 cm3 and their geographic locations were distributed in both lungs. 6MV, 6 MV-FFF, 10 MV, or 10 MV-FFF energies were used. RTOG-0915 metrics conformity index, homogeneity index, D2cm, R50%, and HDloc were evaluated. DFG was computed from the mean and maximum dose in seven concentric 5 mm wide rings outside the PTV. DFG was investigated against the volume of normal lung irradiated by 50% isodose volume. Treatment plans with alternate energy and couch rotations were generated. RESULTS: The dose falloff beyond PTV was modeled using a double exponential fit and evaluated for relationship with intermediate lung dose. Photon energy and beam configuration had a minimal impact on the dose falloff outside. The product of normalized D2cm and R50% was estimated to have a slowly varying value. CONCLUSIONS: Dose falloff outside PTV has been studied as a function of radial distance and ascertained by intermediate dose to normal lung. DFG can serve as a complementary plan quality metric.
ABSTRACT
Twenty-four-hour radioactive iodine uptake (RAIU) is commonly used to compute for therapy dose in Graves disease (GD). The purpose of this study is to determine the feasibility of using 4-hour RAIU in calculating the dose by correlating 4-hour with 24-hour RAIU and comparing the actual therapy dose using 24-hour RAIU with the computed dose using 4-hour RAIU. A total of 83 GD patients (71% female, 29% male;18-50 years old), who underwent RAI therapy at USTH Section of Nuclear Medicine, were included. There was a strong and positive correlation between 4-hour and 24-hour RAIU values (r=0.736). Paired t-test did not show a statistical difference between the actual given therapy dose based on 24-hour RAIU and the computed therapy dose using 4-hour RAIU (p 0.078). This study showed that therapy dose calculation using 4-hour RAIU can be used in Graves disease patients with no rapid turnover.