Applications of Machine Learning to Improve the Clinical Viability of Compton Camera Based in vivo Range Verification in Proton Radiotherapy.

We studied the application of a deep, fully connected Neural Network (NN) to process prompt gamma (PG) data measured by a Compton camera (CC) during the delivery of clinical proton radiotherapy beams. The network identifies 1) recorded "bad" PG events arising from background noise during the measurement, and 2) the correct ordering of PG interactions in the CC to help improve the fidelity of "good" data used for image reconstruction. PG emission from a tissue-equivalent target during irradiation with a 150 MeV proton beam delivered at clinical dose rates was measured with a prototype CC. Images were reconstructed from both the raw measured data and the measured data that was further processed with a neural network (NN) trained to identify "good" and "bad" PG events and predict the ordering of individual interactions within the good PG events. We determine if NN processing of the CC data could improve the reconstructed PG images to a level in which they could provide clinically useful information about the in vivo range and range shifts of the proton beams delivered at full clinical dose rates. Results showed that a deep, fully connected NN improved the achievable contrast to noise ratio (CNR) in our images by more than a factor of 8x. This allowed the path, range, and lateral width of the clinical proton beam within a tissue equivalent target to easily be identified from the PG images, even at the highest dose rates of a 150 MeV proton beam used for clinical treatments. On average, shifts in the beam range as small as 3 mm could be identified. However, when limited by the amount of PG data measured with our prototype CC during the delivery of a single proton pencil beam (~1 × 109 protons), the uncertainty in the reconstructed PG images limited the identification of range shift to ~5 mm. Substantial improvements in CC images were obtained during clinical beam delivery through NN pre-processing of the measured PG data. We believe this shows the potential of NNs to help improve and push CC-based PG imaging toward eventual clinical application for proton RT treatment delivery verification.

Computed Tomography Radiomics Kinetics as Early Imaging Correlates of Osteoradionecrosis in Oropharyngeal Cancer Patients.

Osteoradionecrosis (ORN) is a major side-effect of radiation therapy in oropharyngeal cancer (OPC) patients. In this study, we demonstrate that early prediction of ORN is possible by analyzing the temporal evolution of mandibular subvolumes receiving radiation. For our analysis, we use computed tomography (CT) scans from 21 OPC patients treated with Intensity Modulated Radiation Therapy (IMRT) with subsequent radiographically-proven ≥ grade II ORN, at three different time points: pre-IMRT, 2-months, and 6-months post-IMRT. For each patient, radiomic features were extracted from a mandibular subvolume that developed ORN and a control subvolume that received the same dose but did not develop ORN. We used a Multivariate Functional Principal Component Analysis (MFPCA) approach to characterize the temporal trajectories of these features. The proposed MFPCA model performs the best at classifying ORN vs. Control subvolumes with an area under curve (AUC) = 0.74 [95% confidence interval (C.I.): 0.61-0.90], significantly outperforming existing approaches such as a pre-IMRT features model or a delta model based on changes at intermediate time points, i.e., at 2- and 6-month follow-up. This suggests that temporal trajectories of radiomics features derived from sequential pre- and post-RT CT scans can provide markers that are correlates of RT-induced mandibular injury, and consequently aid in earlier management of ORN.

Use of uniform shots for robust planning of mask-based treatment in Gamma Knife Icon.

PURPOSE: To verify whether Icon automatic correction is robust in preserving plan quality. MATERIALS/METHODS: An end-to-end phantom was used to verify Icon's correction accuracy qualitatively. For quantitative assessment, two plans, a composite- and a uniform-shot-only, were created for an elliptical- (E) and a sausage-shaped (S) lesion inside a PseudoPatient head phantom with a film insert. The phantom was irradiated in the planned and three other positions under each plan: 14° pitch (B); 14° rotation + 8° pitch (C); 95° rotation + 4-cm shift (D). RESULTS: Icon accurately corrects the locations of the shots. For the uniform-shot plans: all gamma index passing rates were >97%, and the differences between the planned and the delivery doses (minimum, maximum, and mean) were all ≤0.1 Gy. For the composite-shot plans, however, the dose differences increased as the phantom was shifted through positions B-D, with a gamma index passing rate of 61% for lesion-E in position D, and 92%, 79%, and 45% for lesion-S in positions B, C, and D, respectively. CONCLUSIONS: Plans using only uniform shots are more robust to deviations in treatment position. The tolerance for such deviations may be lower for plans using composite shots.

Technical Note: Proof of concept for radiomics-based quality assurance for computed tomography.

PURPOSE: Routine quality assurance (QA) testing to identify malfunctions in medical imaging devices is a standard practice and plays an important role in meeting quality standards. However, current daily computed tomography (CT) QA techniques have proven to be inadequate for the detection of subtle artifacts on scans. Therefore, we investigated the ability of a radiomics phantom to detect subtle artifacts not detected in conventional daily QA. METHODS: An updated credence cartridge radiomics phantom was used in this study, with a focus on two of the cartridges (rubber and cork) in the phantom. The phantom was scanned using a Siemens Definition Flash CT scanner, which was reported to produce a subtle line pattern artifact. Images were then imported into the IBEX software program, and 49 features were extracted from the two cartridges using four different preprocessing techniques. Each feature was then compared with features for the same scanner several months previously and with features from controlled CT scans obtained using 100 scanners. RESULTS: Of 196 total features for the test scanner, 79 (40%) from the rubber cartridge and 70 (36%) from the cork cartridge were three or more standard deviations away from the mean of the controlled scan population data. Feature values for the artifact-producing scanner were closer to the population mean when features were preprocessed with Butterworth smoothing. The feature most sensitive to the artifact was co-occurrence matrix maximum probability. The deviation from the mean for this feature was more than seven times greater when the scanner was malfunctioning (7.56 versus 1.01). CONCLUSIONS: Radiomics features extracted from a texture phantom were able to identify an artifact-producing scanner as an outlier among 100 CT scanners. This preliminary analysis demonstrated the potential of radiomics in CT QA to identify subtle artifacts not detected using the currently employed daily QA techniques.

Imaging-Genomic Study of Head and Neck Squamous Cell Carcinoma: Associations Between Radiomic Phenotypes and Genomic Mechanisms via Integration of The Cancer Genome Atlas and The Cancer Imaging Archive.

PURPOSE: Recent data suggest that imaging radiomic features of a tumor could be indicative of important genomic biomarkers. Understanding the relationship between radiomic and genomic features is important for basic cancer research and future patient care. We performed a comprehensive study to discover the imaginggenomic associations in head and neck squamous cell carcinoma (HNSCC) and explore the potential of predicting tumor genomic alternations using radiomic features. METHODS: Our retrospective study integrated whole-genome multiomics data from The Cancer Genome Atlas with matched computed tomography imaging data from The Cancer Imaging Archive for the same set of 126 patients with HNSCC. Linear regression and gene set enrichment analysis were used to identify statistically significant associations between radiomic imaging and genomic features. Random forest classifier was used to predict the status of two key HNSCC molecular biomarkers, human papillomavirus and disruptive TP53 mutation, on the basis of radiomic features. RESULTS: Widespread and statistically significant associations were discovered between genomic features (including microRNA expression, somatic mutations, and transcriptional activity, copy number variations, and promoter region DNA methylation changes of pathways) and radiomic features characterizing the size, shape, and texture of tumor. Prediction of human papillomavirus and TP53 mutation status using radiomic features achieved areas under the receiver operating characteristic curve of 0.71 and 0.641, respectively. CONCLUSION: Our exploratory study suggests that radiomic features are associated with genomic characteristics at multiple molecular layers in HNSCC and provides justification for continued development of radiomics as biomarkers for relevant genomic alterations in HNSCC.

Lung tumor segmentation methods: Impact on the uncertainty of radiomics features for non-small cell lung cancer.

PURPOSE: To evaluate the uncertainty of radiomics features from contrast-enhanced breath-hold helical CT scans of non-small cell lung cancer for both manual and semi-automatic segmentation due to intra-observer, inter-observer, and inter-software reliability. METHODS: Three radiation oncologists manually delineated lung tumors twice from 10 CT scans using two software tools (3D-Slicer and MIM Maestro). Additionally, three observers without formal clinical training were instructed to use two semi-automatic segmentation tools, Lesion Sizing Toolkit (LSTK) and GrowCut, to delineate the same tumor volumes. The accuracy of the semi-automatic contours was assessed by comparison with physician manual contours using Dice similarity coefficients and Hausdorff distances. Eighty-three radiomics features were calculated for each delineated tumor contour. Informative features were identified based on their dynamic range and correlation to other features. Feature reliability was then evaluated using intra-class correlation coefficients (ICC). Feature range was used to evaluate the uncertainty of the segmentation methods. RESULTS: From the initial set of 83 features, 40 radiomics features were found to be informative, and these 40 features were used in the subsequent analyses. For both intra-observer and inter-observer reliability, LSTK had higher reliability than GrowCut and the two manual segmentation tools. All observers achieved consistently high ICC values when using LSTK, but the ICC value varied greatly for each observer when using GrowCut and the manual segmentation tools. For inter-software reliability, features were not reproducible across the software tools for either manual or semi-automatic segmentation methods. Additionally, no feature category was found to be more reproducible than another feature category. Feature ranges of LSTK contours were smaller than those of manual contours for all features. CONCLUSION: Radiomics features extracted from LSTK contours were highly reliable across and among observers. With semi-automatic segmentation tools, observers without formal clinical training were comparable to physicians in evaluating tumor segmentation.

Machine Learning Applications in Head and Neck Radiation Oncology: Lessons From Open-Source Radiomics Challenges.

Radiomics leverages existing image datasets to provide non-visible data extraction via image post-processing, with the aim of identifying prognostic, and predictive imaging features at a sub-region of interest level. However, the application of radiomics is hampered by several challenges such as lack of image acquisition/analysis method standardization, impeding generalizability. As of yet, radiomics remains intriguing, but not clinically validated. We aimed to test the feasibility of a non-custom-constructed platform for disseminating existing large, standardized databases across institutions for promoting radiomics studies. Hence, University of Texas MD Anderson Cancer Center organized two public radiomics challenges in head and neck radiation oncology domain. This was done in conjunction with MICCAI 2016 satellite symposium using Kaggle-in-Class, a machine-learning and predictive analytics platform. We drew on clinical data matched to radiomics data derived from diagnostic contrast-enhanced computed tomography (CECT) images in a dataset of 315 patients with oropharyngeal cancer. Contestants were tasked to develop models for (i) classifying patients according to their human papillomavirus status, or (ii) predicting local tumor recurrence, following radiotherapy. Data were split into training, and test sets. Seventeen teams from various professional domains participated in one or both of the challenges. This review paper was based on the contestants' feedback; provided by 8 contestants only (47%). Six contestants (75%) incorporated extracted radiomics features into their predictive model building, either alone (n = 5; 62.5%), as was the case with the winner of the "HPV" challenge, or in conjunction with matched clinical attributes (n = 2; 25%). Only 23% of contestants, notably, including the winner of the "local recurrence" challenge, built their model relying solely on clinical data. In addition to the value of the integration of machine learning into clinical decision-making, our experience sheds light on challenges in sharing and directing existing datasets toward clinical applications of radiomics, including hyper-dimensionality of the clinical/imaging data attributes. Our experience may help guide researchers to create a framework for sharing and reuse of already published data that we believe will ultimately accelerate the pace of clinical applications of radiomics; both in challenge or clinical settings.

Practical guidelines for handling head and neck computed tomography artifacts for quantitative image analysis.

Radiomics studies have demonstrated the potential use of quantitative image features to improve prognostic stratification of patients with head and neck cancer. Imaging protocol parameters that can affect radiomics feature values have been investigated, but the effects of artifacts caused by intrinsic patient factors have not. Two such artifacts that are common in patients with head and neck cancer are streak artifacts caused by dental fillings and beam-hardening artifacts caused by bone. The purpose of this study was to test the impact of these artifacts and if needed, methods for compensating for these artifacts in head and neck radiomics studies. The robustness of feature values was tested by removing slices of the gross tumor volume (GTV) on computed tomography images from 30 patients with head and neck cancer; these images did not have streak artifacts or had artifacts far from the GTV. The range of each feature value over a percentage of the GTV was compared to the inter-patient variability at full volume. To determine the effects of beam-hardening artifacts, we scanned a phantom with 5 cartridges of different materials encased in polystyrene buildup. A cylindrical hole through the cartridges contained either a rod of polylactic acid to simulate water or a rod of polyvinyl chloride to simulate bone. A region of interest was drawn in each cartridge flush with the rod. Most features were robust with up to 50% of the original GTV removed. Most feature values did not significantly differ when measured with the polylactic acid rod or the polyvinyl chloride rod. Of those that did, the size of the difference did not exceed the inter-patient standard deviation in most cases. We conclude that simply removing slices affected by streak artifacts can enable these scans to be included in radiomics studies and that contours of structures can abut bone without being affected by beam hardening if needed.

Comprehensive Investigation on Controlling for CT Imaging Variabilities in Radiomics Studies.

Radiomics has shown promise in improving models for predicting patient outcomes. However, to maximize the information gain of the radiomics features, especially in larger patient cohorts, the variability in radiomics features owing to differences between scanners and scanning protocols must be accounted for. To this aim, the imaging variability of radiomics feature values was evaluated on 100 computed tomography scanners at 35 clinics by imaging a radiomics phantom using a controlled protocol and the commonly used chest and head protocols of the local clinic. We used a linear mixed-effects model to determine the degree to which the manufacturer and individual scanners contribute to the overall variability. Using a controlled protocol reduced the overall variability by 57% and 52% compared to the local chest and head protocols respectively. The controlled protocol also reduced the relative contribution of the manufacturer to the total variability. For almost all variabilities (manufacturer, scanner, and residual with different preprocesssing), the controlled protocol scans had a significantly smaller variability than the local protocol scans did. For most radiomics features, the imaging variability was small relative to the inter-patient feature variability in non-small cell lung cancer and head and neck squamous cell carcinoma patient cohorts. From this study, we conclude that using controlled scans can reduce the variability in radiomics features, and our results demonstrate the importance of using controlled protocols in prospective radiomics studies.

Guidelines and Experience Using Imaging Biomarker Explorer (IBEX) for Radiomics.

Imaging Biomarker Explorer (IBEX) is an open-source tool for medical imaging radiomics work. The purpose of this paper is to describe how to use IBEX's graphical user interface (GUI) and to demonstrate how IBEX calculated features have been used in clinical studies. IBEX allows for the import of DICOM images with DICOM radiation therapy structure files or Pinnacle files. Once the images are imported, IBEX has tools within the Data Selection GUI to manipulate the viewing of the images, measure voxel values and distances, and create and edit contours. IBEX comes with 27 preprocessing and 132 feature choices to design feature sets. Each preprocessing and feature category has parameters that can be altered. The output from IBEX is a spreadsheet that contains: 1) each feature from the feature set calculated for each contour in a data set, 2) image information about each contour in a data set, and 3) a summary of the preprocessing and features used with their selected parameters. Features calculated from IBEX have been used in studies to test the variability of features under different imaging conditions and in survival models to improve current clinical models.

Prognostic value of combining a quantitative image feature from positron emission tomography with clinical factors in oligometastatic non-small cell lung cancer.

BACKGROUND AND PURPOSE: Oligometastatic non-small cell lung cancer (NSCLC) is a heterogeneous condition with few known risk stratification factors. A quantitative imaging feature (QIF) on positron emission tomography (PET), gray-level co-occurrence matrix energy, has been linked with outcome of nonmetastatic NSCLC. We hypothesized that GLCM energy would enhance the ability of models comprising standard clinical prognostic factors (CPFs) to stratify oligometastatic patients based on overall survival (OS). MATERIALS AND METHODS: We assessed 79 patients with oligometastatic NSCLC (≤3 metastases) diagnosed in 2007-2015. The primary and largest metastases at diagnosis were delineated on pretreatment scans with GLCM energy extracted using imaging biomarker explorer (IBEX) software. Iterative stepwise elimination feature selection based on the Akaike information criterion identified the optimal model comprising CPFs for predicting OS in a multivariate Cox proportional hazards model. GLCM energy was tested for improving prediction accuracy. RESULTS: Energy was a significant predictor of OS (P = 0.028) in addition to the selected CPFs. The c-indexes for the CPF-only and CPF + Energy models were 0.720 and 0.739. CONCLUSIONS: Incorporating Energy strengthened a CPF model for predicting OS. These findings support further exploration of QIFs, including markers of the primary tumor vs. those of the metastatic sites.

Cost-effective immobilization for whole brain radiation therapy.

To investigate the inter- and intra-fraction motion associated with the use of a low-cost tape immobilization technique as an alternative to thermoplastic immobilization masks for whole-brain treatments. The results of this study may be of interest to clinical staff with severely limited resources (e.g., in low-income countries) and also when treating patients who cannot tolerate standard immobilization masks. Setup reproducibility of eight healthy volunteers was assessed for two different immobilization techniques. (a) One strip of tape was placed across the volunteer's forehead and attached to the sides of the treatment table. (b) A second strip was added to the first, under the chin, and secured to the table above the volunteer's head. After initial positioning, anterior and lateral photographs were acquired. Volunteers were positioned five times with each technique to allow calculation of inter-fraction reproducibility measurements. To estimate intra-fraction reproducibility, 5-minute anterior and lateral videos were taken for each technique per volunteer. An in-house software was used to analyze the photos and videos to assess setup reproducibility. The maximum intra-fraction displacement for all volunteers was 2.8 mm. Intra-fraction motion increased with time on table. The maximum inter-fraction range of positions for all volunteers was 5.4 mm. The magnitude of inter-fraction and intra-fraction motion found using the "1-strip" and "2-strip" tape immobilization techniques was comparable to motion restrictions provided by a thermoplastic mask for whole-brain radiotherapy. The results suggest that tape-based immobilization techniques represent an economical and useful alternative to the thermoplastic mask.

Reply to Comment on 'Imaging of prompt gamma rays emitted during delivery of clinical proton beams with a Compton camera: feasibility studies for range verification'.

A reply is provided to the points raised in the comment by Dr Sitek (2016 Phys. Med. Biol. 61 8941) on Polf et al (2015 Phys. Med. Biol. 60 7085).

Quantitative analysis of treatment process time and throughput capacity for spot scanning proton therapy.

PURPOSE: To determine the patient throughput and the overall efficiency of the spot scanning system by analyzing treatment time, equipment availability, and maximum daily capacity for the current spot scanning port at Proton Therapy Center Houston and to assess the daily throughput capacity for a hypothetical spot scanning proton therapy center. METHODS: At their proton therapy center, the authors have been recording in an electronic medical record system all treatment data, including disease site, number of fields, number of fractions, delivered dose, energy, range, number of spots, and number of layers for every treatment field. The authors analyzed delivery system downtimes that had been recorded for every equipment failure and associated incidents. These data were used to evaluate the patient census, patient distribution as a function of the number of fields and total target volume, and equipment clinical availability. The duration of each treatment session from patient walk-in to patient walk-out of the spot scanning treatment room was measured for 64 patients with head and neck, central nervous system, thoracic, and genitourinary cancers. The authors retrieved data for total target volume and the numbers of layers and spots for all fields from treatment plans for a total of 271 patients (including the above 64 patients). A sensitivity analysis of daily throughput capacity was performed by varying seven parameters in a throughput capacity model. RESULTS: The mean monthly equipment clinical availability for the spot scanning port in April 2012-March 2015 was 98.5%. Approximately 1500 patients had received spot scanning proton therapy as of March 2015. The major disease sites treated in September 2012-August 2014 were the genitourinary system (34%), head and neck (30%), central nervous system (21%), and thorax (14%), with other sites accounting for the remaining 1%. Spot scanning beam delivery time increased with total target volume and accounted for approximately 30%-40% of total treatment time for the total target volumes exceeding 200 cm(3), which was the case for more than 80% of the patients in this study. When total treatment time was modeled as a function of the number of fields and total target volume, the model overestimated total treatment time by 12% on average, with a standard deviation of 32%. A sensitivity analysis of throughput capacity for a hypothetical four-room spot scanning proton therapy center identified several priority items for improvements in throughput capacity, including operation time, beam delivery time, and patient immobilization and setup time. CONCLUSIONS: The spot scanning port at our proton therapy center has operated at a high performance level and has been used to treat a large number of complex cases. Further improvements in efficiency may be feasible in the areas of facility operation, beam delivery, patient immobilization and setup, and optimization of treatment scheduling.

Imaging of prompt gamma rays emitted during delivery of clinical proton beams with a Compton camera: feasibility studies for range verification.

The purpose of this paper is to evaluate the ability of a prototype Compton camera (CC) to measure prompt gamma rays (PG) emitted during delivery of clinical proton pencil beams for prompt gamma imaging (PGI) as a means of providing in vivo verification of the delivered proton radiotherapy beams. A water phantom was irradiated with clinical 114 MeV and 150 MeV proton pencil beams. Up to 500 cGy of dose was delivered per irradiation using clinical beam currents. The prototype CC was placed 15 cm from the beam central axis and PGs from 0.2 MeV up to 6.5 MeV were measured during irradiation. From the measured data (2D) images of the PG emission were reconstructed. (1D) profiles were extracted from the PG images and compared to measured depth dose curves of the delivered proton pencil beams. The CC was able to measure PG emission during delivery of both 114 MeV and 150 MeV proton beams at clinical beam currents. 2D images of the PG emission were reconstructed for single 150 MeV proton pencil beams as well as for a 5 × 5 cm mono-energetic layer of 114 MeV pencil beams. Shifts in the Bragg peak (BP) range were detectable on the 2D images. 1D profiles extracted from the PG images show that the distal falloff of the PG emission profile lined up well with the distal BP falloff. Shifts as small as 3 mm in the beam range could be detected from the 1D PG profiles with an accuracy of 1.5 mm or better. However, with the current CC prototype, a dose of 400 cGy was required to acquire adequate PG signal for 2D PG image reconstruction. It was possible to measure PG interactions with our prototype CC during delivery of proton pencil beams at clinical dose rates. Images of the PG emission could be reconstructed and shifts in the BP range were detectable. Therefore PGI with a CC for in vivo range verification during proton treatment delivery is feasible. However, improvements in the prototype CC detection efficiency and reconstruction algorithms are necessary to make it a clinically viable PGI system.

Measurement of characteristic prompt gamma rays emitted from oxygen and carbon in tissue-equivalent samples during proton beam irradiation.

The purpose of this work was to characterize how prompt gamma (PG) emission from tissue changes as a function of carbon and oxygen concentration, and to assess the feasibility of determining elemental concentration in tissues irradiated with proton beams. For this study, four tissue-equivalent water-sucrose samples with differing densities and concentrations of carbon, hydrogen, and oxygen were irradiated with a 48 MeV proton pencil beam. The PG spectrum emitted from each sample was measured using a high-purity germanium detector, and the absolute detection efficiency of the detector, average beam current, and delivered dose distribution were also measured. Changes to the total PG emission from (12)C (4.44 MeV) and (16)O (6.13 MeV) per incident proton and per Gray of absorbed dose were characterized as a function of carbon and oxygen concentration in the sample. The intensity of the 4.44 MeV PG emission per incident proton was found to be nearly constant for all samples regardless of their carbon concentration. However, we found that the 6.13 MeV PG emission increased linearly with the total amount (in grams) of oxygen irradiated in the sample. From the measured PG data, we determined that 1.64 × 10(7) oxygen PGs were emitted per gram of oxygen irradiated per Gray of absorbed dose delivered with a 48 MeV proton beam. These results indicate that the 6.13 MeV PG emission from (16)O is proportional to the concentration of oxygen in tissue irradiated with proton beams, showing that it is possible to determine the concentration of oxygen within tissues irradiated with proton beams by measuring (16)O PG emission.