Development and deployment of a histopathology-based deep learning algorithm for patient prescreening in a clinical trial.

Accurate identification of genetic alterations in tumors, such as Fibroblast Growth Factor Receptor, is crucial for treating with targeted therapies; however, molecular testing can delay patient care due to the time and tissue required. Successful development, validation, and deployment of an AI-based, biomarker-detection algorithm could reduce screening cost and accelerate patient recruitment. Here, we develop a deep-learning algorithm using >3000 H&E-stained whole slide images from patients with advanced urothelial cancers, optimized for high sensitivity to avoid ruling out trial-eligible patients. The algorithm is validated on a dataset of 350 patients, achieving an area under the curve of 0.75, specificity of 31.8% at 88.7% sensitivity, and projected 28.7% reduction in molecular testing. We successfully deploy the system in a non-interventional study comprising 89 global study clinical sites and demonstrate its potential to prioritize/deprioritize molecular testing resources and provide substantial cost savings in the drug development and clinical settings.

Accelerating pharmaceutical R&D with a user-friendly AI system for histopathology image analysis.

A system for analysis of histopathology data within a pharmaceutical R&D environment has been developed with the intention of enabling interdisciplinary collaboration. State-of-the-art AI tools have been deployed as easy-to-use self-service modules within an open-source whole slide image viewing platform, so that non-data scientist users (e.g., clinicians) can utilize and evaluate pre-trained algorithms and retrieve quantitative results. The outputs of analysis are automatically cataloged in the database to track data provenance and can be viewed interactively on the slide as annotations or heatmaps. Commonly used models for analysis of whole slide images including segmentation, extraction of hand-engineered features for segmented regions, and slide-level classification using multi-instance learning are included and new models can be added as needed. The source code that supports running inference with these models internally is backed up by a robust CI/CD pipeline to ensure model versioning, robust testing, and seamless deployment of the latest models. Examples of the use of this system in a pharmaceutical development workflow include glomeruli segmentation, enumeration of podocyte count from WT-1 immuno-histochemistry, measurement of beta-1 integrin target engagement from immunofluorescence, digital glomerular phenotyping from periodic acid-Schiff histology, PD-L1 score prediction using multi-instance learning, and the deployment of the open-source Segment Anything model to speed up annotation.

Development and validation of a longitudinal soft-tissue metastatic lesion matching algorithm.

Metastatic cancer presents with many, sometimes hundreds of metastatic lesions through the body, which often respond heterogeneously to treatment. Therefore, lesion-level assessment is necessary for a complete understanding of disease response. Lesion-level assessment typically requires manual matching of corresponding lesions, which is a tedious, subjective, and error-prone task. This study introduces a fully automated algorithm for matching of metastatic lesions in longitudinal medical images. The algorithm entails four steps: (1) image registration, (2) lesion dilation, (3) lesion clustering, and (4) linear assignment. In step (1), 3D deformable registration is used to register the scans. In step (2), lesion contours are conformally dilated. In step (3), lesion clustering is evaluated based on local metrics. In step (4), matching is assigned based on non-greedy cost minimization. The algorithm was optimized (e.g. choice of deformable registration algorithm, dilatation size) and validated on 140 scan-pairs of 32 metastatic cancer patients from two independent clinical trials, who received longitudinal PET/CT scans as part of their treatment response assessment. Registration error was evaluated using landmark distance. A sensitivity study was performed to evaluate the optimal lesion dilation magnitude. Lesion matching performance accuracy was evaluated for all patients and for a subset with high disease burden. Two investigated deformable registration approaches (whole body deformable and articulated deformable registrations) led to similar performance with the overall registration accuracy between 2.3 and 2.6 mm. The optimal dilation magnitude of 25 mm yielded almost a perfect matching accuracy of 0.98. No significant matching accuracy decrease was observed in the subset of patients with high lesion disease burden. In summary, lesion matching using our new algorithm was highly accurate and a significant improvement, when compared to previously established methods. The proposed method enables accurate automated metastatic lesion matching in whole-body longitudinal scans.

Prognostic Value of Computed Tomography and/or 18F-Fluorodeoxyglucose Positron Emission Tomography Radiomics Features in Locally Advanced Non-small Cell Lung Cancer.

INTRODUCTION: We investigated whether adding computed tomography (CT) and/or 18F-fluorodeoxyglucose (18F-FDG) PET radiomics features to conventional prognostic factors (CPFs) improves prognostic value in locally advanced non-small cell lung cancer (NSCLC). MATERIALS AND METHODS: We retrospectively identified 39 cases with stage III NSCLC who received chemoradiotherapy and underwent planning CT and staging 18F-FDG PET scans. Seven CPFs were recorded. Feature selection was performed on 48 CT and 49 PET extracted radiomics features. A penalized multivariate Cox proportional hazards model was used to generate models for overall survival based on CPFs alone, CPFs with CT features, CPFs with PET features, and CPFs with CT and PET features. Linear predictors generated and categorized into 2 risk groups for which Kaplan-Meier survival curves were calculated. A log-rank test was performed to quantify the discrimination between the groups and calculated the Harrell's C-index to quantify the discriminatory power. A likelihood ratio test was performed to determine whether adding CT and/or PET features to CPFs improved model performance. RESULTS: All 4 models significantly discriminated between the 2 risk groups. The discriminatory power was significantly increased when CPFs were combined with PET features (C-index 0.82; likelihood ratio test P < .01) or with both CT and PET features (0.83; P < .01) compared with CPFs alone (0.68). There was no significant improvement when CPFs were combined with CT features (0.68). CONCLUSION: Adding PET radiomics features to CPFs yielded a significant improvement in the prognostic value in locally advanced NSCLC; adding CT features did not.

Machine Learning-Based Multiparametric Magnetic Resonance Imaging Radiomics for Prediction of H3K27M Mutation in Midline Gliomas.

OBJECTIVE: H3K27M mutation in gliomas has prognostic implications. Previous magnetic resonance imaging (MRI) studies have reported variable rates of tumoral enhancement, necrotic changes, and peritumoral edema in H3K27M-mutant gliomas, with no distinguishing imaging features compared with wild-type gliomas. We aimed to construct an MRI machine learning (ML)-based radiomic model to predict H3K27M mutation in midline gliomas. METHODS: A total of 109 patients from 3 academic centers were included in this study. Fifty patients had H3K27M mutation and 59 were wild-type. Conventional MRI sequences (T1-weighted, T2-weighted, T2-fluid-attenuated inversion recovery, postcontrast T1-weighted, and apparent diffusion coefficient maps) were used for feature extraction. A total of 651 radiomic features per each sequence were extracted. Patients were randomly selected with a 7:3 ratio to create training (n = 76) and test (n = 33) data sets. An extreme gradient boosting algorithm (XGBoost) was used in ML-based model development. Performance of the model was assessed by area under the receiver operating characteristic curve. RESULTS: Pediatric patients accounted for a larger proportion of the study cohort (60 pediatric [55%] vs. 49 adult [45%] patients). XGBoost with additional feature selection had an area under the receiver operating characteristic curve of 0.791 and 0.737 in the training and test data sets, respectively. The model achieved accuracy, precision (positive predictive value), recall (sensitivity), and F1 (harmonic mean of precision and recall) measures of 72.7%, 76.5%, 72.2%, and 74.3%, respectively, in the test set. CONCLUSIONS: Our multi-institutional study suggests that ML-based radiomic analysis of multiparametric MRI can be a promising noninvasive technique to predict H3K27M mutation status in midline gliomas.

Texture analysis of T2-weighted MRI predicts SDH mutation in paraganglioma.

PURPOSE: Texture analysis can quantify sophisticated imaging characteristics. We hypothesized that 2D textures computed with T2-weighted and post-contrast T1-weighted MRI can predict succinate dehydrogenase (SDH) mutation status in head and neck paragangliomas. METHODS: Our retrospective study included 21 patients (1 to 4 tumors/patient) with 24 pathologically proven paragangliomas in the head and neck. Fourteen lesions (58%) were SDH mutation-positive. All patients underwent T2-weighted and post-contrast T1-weighted MRI sequences. Three 2D texture features of dependence non-uniformity normalized (DNN), small dependence high gray level emphasis (SDHGLE), and small dependence low gray level emphasis (SDLGLE) were calculated. Computed textures between SDH mutants and non-mutants were compared using Mann-Whitney U test. Area under the receiver operating characteristic (AUROC) curve was used to quantify the predictive power of each texture. RESULTS: Only T2-based SDLGLE was statistically significant (p = 0.048), and AUROC was 0.71. Diagnostic accuracy was 70.8%. CONCLUSION: 2D texture parameter of T2-based SDLGLE predicts SDH mutation in head and neck paragangliomas. This noninvasive technique can potentially facilitate further genetic workup.

Multi-Field-of-View Deep Learning Model Predicts Nonsmall Cell Lung Cancer Programmed Death-Ligand 1 Status from Whole-Slide Hematoxylin and Eosin Images.

BACKGROUND: Tumor programmed death-ligand 1 (PD-L1) status is useful in determining which patients may benefit from programmed death-1 (PD-1)/PD-L1 inhibitors. However, little is known about the association between PD-L1 status and tumor histopathological patterns. Using deep learning, we predicted PD-L1 status from hematoxylin and eosin (H and E) whole-slide images (WSIs) of nonsmall cell lung cancer (NSCLC) tumor samples. MATERIALS AND METHODS: One hundred and thirty NSCLC patients were randomly assigned to training (n = 48) or test (n = 82) cohorts. A pair of H and E and PD-L1-immunostained WSIs was obtained for each patient. A pathologist annotated PD-L1 positive and negative tumor regions on the training samples using immunostained WSIs for reference. From the H and E WSIs, over 145,000 training tiles were generated and used to train a multi-field-of-view deep learning model with a residual neural network backbone. RESULTS: The trained model accurately predicted tumor PD-L1 status on the held-out test cohort of H and E WSIs, which was balanced for PD-L1 status (area under the receiver operating characteristic curve [AUC] =0.80, P << 0.01). The model remained effective over a range of PD-L1 cutoff thresholds (AUC = 0.67-0.81, P ≤ 0.01) and when different proportions of the labels were randomly shuffled to simulate interpathologist disagreement (AUC = 0.63-0.77, P ≤ 0.03). CONCLUSIONS: A robust deep learning model was developed to predict tumor PD-L1 status from H and E WSIs in NSCLC. These results suggest that PD-L1 expression is correlated with the morphological features of the tumor microenvironment.

Semi-automated pulmonary nodule interval segmentation using the NLST data.

PURPOSE: To study the variability in volume change estimates of pulmonary nodules due to segmentation approaches used across several algorithms and to evaluate these effects on the ability to predict nodule malignancy. METHODS: We obtained 100 patient image datasets from the National Lung Screening Trial (NLST) that had a nodule detected on each of two consecutive low dose computed tomography (LDCT) scans, with an equal proportion of malignant and benign cases (50 malignant, 50 benign). Information about the nodule location for the cases was provided by a screen capture with a bounding box and its axial location was indicated. Five participating quantitative imaging network (QIN) institutions performed nodule segmentation using their preferred semi-automated algorithms with no manual correction; teams were allowed to provide additional manually corrected segmentations (analyzed separately). The teams were asked to provide segmentation masks for each nodule at both time points. From these masks, the volume was estimated for the nodule at each time point; the change in volume (absolute and percent change) across time points was estimated as well. We used the concordance correlation coefficient (CCC) to compare the similarity of computed nodule volumes (absolute and percent change) across algorithms. We used Logistic regression model on the change in volume (absolute change and percent change) of the nodules to predict the malignancy status, the area under the receiver operating characteristic curve (AUROC) and confidence intervals were reported. Because the size of nodules was expected to have a substantial effect on segmentation variability, analysis of change in volumes was stratified by lesion size, where lesions were grouped into those with a longest diameter of <8 mm and those with longest diameter ≥ 8 mm. RESULTS: We find that segmentation of the nodules shows substantial variability across algorithms, with the CCC ranging from 0.56 to 0.95 for change in volume (percent change in volume range was [0.15 to 0.86]) across the nodules. When examining nodules based on their longest diameter, we find the CCC had higher values for large nodules with a range of [0.54 to 0.93] among the algorithms, while percent change in volume was [0.3 to 0.95]. Compared to that of smaller nodules which had a range of [-0.0038 to 0.69] and percent change in volume was [-0.039 to 0.92]. The malignancy prediction results showed fairly consistent results across the institutions, the AUC using change in volume ranged from 0.65 to 0.89 (Percent change in volume was 0.64 to 0.86) for entire nodule range. Prediction improves for large nodule range (≥ 8 mm) with AUC range 0.75 to 0.90 (percent change in volume was 0.74 to 0.92). Compared to smaller nodule range (<8 mm) with AUC range 0.57 to 0.78 (percent change in volume was 0.59 to 0.77). CONCLUSIONS: We find there is a fairly high concordance in the size measurements for larger nodules (≥8 mm) than the lower sizes (<8 mm) across algorithms. We find the change in nodule volume (absolute and percent change) were consistent predictors of malignancy across institutions, despite using different segmentation algorithms. Using volume change estimates without corrections shows slightly lower predictability (for two teams).

Impact of experimental design on PET radiomics in predicting somatic mutation status.

PURPOSE: PET-based radiomic features have demonstrated great promises in predicting genetic data. However, various experimental parameters can influence the feature extraction pipeline, and hence, Here, we investigated how experimental settings affect the performance of radiomic features in predicting somatic mutation status in non-small cell lung cancer (NSCLC) patients. METHODS: 348 NSCLC patients with somatic mutation testing and diagnostic PET images were included in our analysis. Radiomic feature extractions were analyzed for varying voxel sizes, filters and bin widths. 66 radiomic features were evaluated. The performance of features in predicting mutations status was assessed using the area under the receiver-operating-characteristic curve (AUC). The influence of experimental parameters on feature predictability was quantified as the relative difference between the minimum and maximum AUC (δ). RESULTS: The large majority of features (n=56, 85%) were significantly predictive for EGFR mutation status (AUC≥0.61). 29 radiomic features significantly predicted EGFR mutations and were robust to experimental settings with δOverall<5%. The overall influence (δOverall) of the voxel size, filter and bin width for all features ranged from 5% to 15%, respectively. For all features, none of the experimental designs was predictive of KRAS+ from KRAS- (AUC≤0.56). CONCLUSION: The predictability of 29 radiomic features was robust to the choice of experimental settings; however, these settings need to be carefully chosen for all other features. The combined effect of the investigated processing methods could be substantial and must be considered. Optimized settings that will maximize the predictive performance of individual radiomic features should be investigated in the future.

Radiomics-based Assessment of Radiation-induced Lung Injury After Stereotactic Body Radiotherapy.

BACKGROUND: Over 50% of patients who receive stereotactic body radiotherapy (SBRT) develop radiographic evidence of radiation-induced lung injury. Radiomics is an emerging approach that extracts quantitative features from image data, which may provide greater value and a better understanding of pulmonary toxicity than conventional approaches. We aimed to investigate the potential of computed tomography-based radiomics in characterizing post-SBRT lung injury. METHODS: A total of 48 diagnostic thoracic computed tomography scans (acquired prior to SBRT and at 3, 6, and 9 months post-SBRT) from 14 patients were analyzed. Nine radiomic features (ie, 7 gray level co-occurrence matrix [GLCM] texture features and 2 first-order features) were investigated. The ability of radiomic features to distinguish radiation oncologist-defined moderate/severe lung injury from none/mild lung injury was assessed using logistic regression and area under the receiver operating characteristic curve (AUC). Moreover, dose-response curves (DRCs) for radiomic feature changes were determined as a function of time to investigate whether there was a significant dose-response relationship. RESULTS: The GLCM features (logistic regression P-value range, 0.012-0.262; AUC range, 0.643-0.750) outperformed the first-order features (P-value range, 0.100-0.990; AUC range, 0.543-0.661) in distinguishing lung injury severity levels. Eight of 9 radiomic features demonstrated a significant dose-response relationship at 3, 6, and 9 months post-SBRT. Although not statistically significant, the GLCM features showed clear separations between the 3- or 6-month DRC and the 9-month DRC. CONCLUSION: Radiomic features significantly correlated with radiation oncologist-scored post-SBRT lung injury and showed a significant dose-response relationship, suggesting the potential for radiomics to provide a quantitative, objective measurement of post-SBRT lung injury.

Application of the 3D slicer chest imaging platform segmentation algorithm for large lung nodule delineation.

PURPOSE: Accurate segmentation of lung nodules is crucial in the development of imaging biomarkers for predicting malignancy of the nodules. Manual segmentation is time consuming and affected by inter-observer variability. We evaluated the robustness and accuracy of a publically available semiautomatic segmentation algorithm that is implemented in the 3D Slicer Chest Imaging Platform (CIP) and compared it with the performance of manual segmentation. METHODS: CT images of 354 manually segmented nodules were downloaded from the LIDC database. Four radiologists performed the manual segmentation and assessed various nodule characteristics. The semiautomatic CIP segmentation was initialized using the centroid of the manual segmentations, thereby generating four contours for each nodule. The robustness of both segmentation methods was assessed using the region of uncertainty (δ) and Dice similarity index (DSI). The robustness of the segmentation methods was compared using the Wilcoxon-signed rank test (pWilcoxon<0.05). The Dice similarity index (DSIAgree) between the manual and CIP segmentations was computed to estimate the accuracy of the semiautomatic contours. RESULTS: The median computational time of the CIP segmentation was 10 s. The median CIP and manually segmented volumes were 477 ml and 309 ml, respectively. CIP segmentations were significantly more robust than manual segmentations (median δCIP = 14ml, median dsiCIP = 99% vs. median δmanual = 222ml, median dsimanual = 82%) with pWilcoxon~10-16. The agreement between CIP and manual segmentations had a median DSIAgree of 60%. While 13% (47/354) of the nodules did not require any manual adjustment, minor to substantial manual adjustments were needed for 87% (305/354) of the nodules. CIP segmentations were observed to perform poorly (median DSIAgree≈50%) for non-/sub-solid nodules with subtle appearances and poorly defined boundaries. CONCLUSION: Semi-automatic CIP segmentation can potentially reduce the physician workload for 13% of nodules owing to its computational efficiency and superior stability compared to manual segmentation. Although manual adjustment is needed for many cases, CIP segmentation provides a preliminary contour for physicians as a starting point.

Technical Note: Combination of multiple EPID imager layers improves image quality and tracking performance of low contrast-to-noise objects.

PURPOSE: We hypothesized that combining multiple amorphous silicon flat panel layers increases photon detection efficiency in an electronic portal imaging device (EPID), improving image quality and tracking accuracy of low-contrast targets during radiotherapy. METHODS: The prototype imager evaluated in this study contained four individually programmable layers each with a copper converter layer, Gd2 O2 S scintillator, and active-matrix flat panel imager (AMFPI). The imager was placed on a Varian TrueBeam linac and a Las Vegas phantom programmed with sinusoidal motion (peak-to-peak amplitude = 20 mm, period = 3.5 s) was imaged at a frame rate of 10 Hz with one to four layers activated. Number of visible circles and CNR of least visible circle (depth = 0.5 mm, diameter = 7 mm) was computed to assess the image quality of single and multiple layers. A previously validated tracking algorithm was employed for auto-tracking. Tracking error was defined as the difference between the programmed and tracked positions of the circle. Pearson correlation coefficient (R) of CNR and tracking errors was computed. RESULTS: Motion-induced blurring significantly reduced circle visibility. During four cycles of phantom motion, the number of visible circles varied from 11-23, 13-24, 15-25, and 16-26 for one-, two-, three-, and four-layer imagers, respectively. Compared with using only a single layer, combining two, three, and four layers increased the median CNR by factors of 1.19, 1.42, and 1.71, respectively and reduced the average tracking error from 3.32 mm to 1.67 mm to 1.47 mm, and 0.74 mm, respectively. Significant correlations (P~10-9 ) were found between the tracking error and CNR. CONCLUSION: Combination of four conventional EPID layers significantly improves the EPID image quality and tracking accuracy for a poorly visible object which is moving with a frequency and amplitude similar to respiratory motion.

Associations between radiologist-defined semantic and automatically computed radiomic features in non-small cell lung cancer.

Tumor phenotypes captured in computed tomography (CT) images can be described qualitatively and quantitatively using radiologist-defined "semantic" and computer-derived "radiomic" features, respectively. While both types of features have shown to be promising predictors of prognosis, the association between these groups of features remains unclear. We investigated the associations between semantic and radiomic features in CT images of 258 non-small cell lung adenocarcinomas. The tumor imaging phenotypes were described using 9 qualitative semantic features that were scored by radiologists, and 57 quantitative radiomic features that were automatically calculated using mathematical algorithms. Of the 9 semantic features, 3 were rated on a binary scale (cavitation, air bronchogram, and calcification) and 6 were rated on a categorical scale (texture, border definition, contour, lobulation, spiculation, and concavity). 32-41 radiomic features were associated with the binary semantic features (AUC = 0.56-0.76). The relationship between all radiomic features and the categorical semantic features ranged from weak to moderate (|Spearmen's correlation| = 0.002-0.65). There are associations between semantic and radiomic features, however the associations were not strong despite being significant. Our results indicate that radiomic features may capture distinct tumor phenotypes that fail to be perceived by naked eye that semantic features do not describe and vice versa.

Associations Between Somatic Mutations and Metabolic Imaging Phenotypes in Non-Small Cell Lung Cancer.

PET-based radiomics have been used to noninvasively quantify the metabolic tumor phenotypes; however, little is known about the relationship between these phenotypes and underlying somatic mutations. This study assessed the association and predictive power of 18F-FDG PET-based radiomic features for somatic mutations in non-small cell lung cancer patients. Methods: Three hundred forty-eight non-small cell lung cancer patients underwent diagnostic 18F-FDG PET scans and were tested for genetic mutations. Thirteen percent (44/348) and 28% (96/348) of patients were found to harbor epidermal growth factor receptor (EGFR) or Kristen rat sarcoma viral (KRAS) mutations, respectively. We evaluated 21 imaging features: 19 independent radiomic features quantifying phenotypic traits and 2 conventional features (metabolic tumor volume and maximum SUV). The association between imaging features and mutation status (e.g., EGFR-positive [EGFR+] vs. EGFR-negative) was assessed using the Wilcoxon rank-sum test. The ability of each imaging feature to predict mutation status was evaluated by the area under the receiver operating curve (AUC) and its significance was compared with a random guess (AUC = 0.5) using the Noether test. All P values were corrected for multiple hypothesis testing by controlling the false-discovery rate (FDRWilcoxon, FDRNoether) with a significance threshold of 10%. Results: Eight radiomic features and both conventional features were significantly associated with EGFR mutation status (FDRWilcoxon = 0.01-0.10). One radiomic feature (normalized inverse difference moment) outperformed all other features in predicting EGFR mutation status (EGFR+ vs. EGFR-negative, AUC = 0.67, FDRNoether = 0.0032), as well as differentiating between KRAS-positive and EGFR+ (AUC = 0.65, FDRNoether = 0.05). None of the features was associated with or predictive of KRAS mutation status (KRAS-positive vs. KRAS-negative, AUC = 0.50-0.54). Conclusion: Our results indicate that EGFR mutations may drive different metabolic tumor phenotypes that are captured in PET images, whereas KRAS-mutated tumors do not. This proof-of-concept study sheds light on genotype-phenotype interactions, using radiomics to capture and describe the phenotype, and may have potential for developing noninvasive imaging biomarkers for somatic mutations.

Applications and limitations of radiomics.

Radiomics is an emerging field in quantitative imaging that uses advanced imaging features to objectively and quantitatively describe tumour phenotypes. Radiomic features have recently drawn considerable interest due to its potential predictive power for treatment outcomes and cancer genetics, which may have important applications in personalized medicine. In this technical review, we describe applications and challenges of the radiomic field. We will review radiomic application areas and technical issues, as well as proper practices for the designs of radiomic studies.

Relationship between the Temporal Changes in Positron-Emission-Tomography-Imaging-Based Textural Features and Pathologic Response and Survival in Esophageal Cancer Patients.

PURPOSE: Although change in standardized uptake value (SUV) measures and PET-based textural features during treatment have shown promise in tumor response prediction, it is unclear which quantitative measure is the most predictive. We compared the relationship between PET-based features and pathologic response and overall survival with the SUV measures in esophageal cancer. METHODS: Fifty-four esophageal cancer patients received PET/CT scans before and after chemoradiotherapy. Of these, 45 patients underwent surgery and were classified into complete, partial, and non-responders to the preoperative chemoradiation. SUVmax and SUVmean, two cooccurrence matrix (Entropy and Homogeneity), two run-length matrix (RLM) (high-gray-run emphasis and Short-run high-gray-run emphasis), and two size-zone matrix (high-gray-zone emphasis and short-zone high-gray emphasis) textures were computed. The relationship between the relative difference of each measure at different treatment time points and the pathologic response and overall survival was assessed using the area under the receiver-operating-characteristic curve (AUC) and Kaplan-Meier statistics, respectively. RESULTS: All Textures, except Homogeneity, were better related to pathologic response than SUVmax and SUVmean. Entropy was found to significantly distinguish non-responders from the complete (AUC = 0.79, p = 1.7 × 10(-4)) and partial (AUC = 0.71, p = 0.01) responders. Non-responders can also be significantly differentiated from partial and complete responders by the change in the run-length and size-zone matrix textures (AUC = 0.71-0.76, p ≤ 0.02). Homogeneity, SUVmax, and SUVmean failed to differentiate between any of the responders (AUC = 0.50-0.57, p ≥ 0.46). However, none of the measures were found to significantly distinguish between complete and partial responders with AUC <0.60 (p = 0.37). Median Entropy and RLM textures significantly discriminated patients with good and poor survival (log-rank p < 0.02), while all other textures and survival were poorly related (log-rank p > 0.25). CONCLUSION: For the patients studied, temporal changes in Entropy and all RLM were better correlated with pathological response and survival than the SUV measures. The hypothesis that these metrics can be used as clinical predictors of better patient outcomes will be tested in a larger patient dataset in the future.

Use of registration-based contour propagation in texture analysis for esophageal cancer pathologic response prediction.

Change in PET-based textural features has shown promise in predicting cancer response to treatment. However, contouring tumour volumes on longitudinal scans is time-consuming. This study investigated the usefulness of contour propagation in texture analysis for the purpose of pathologic response prediction in esophageal cancer. Forty-five esophageal cancer patients underwent PET/CT scans before and after chemo-radiotherapy. Patients were classified into responders and non-responders after the surgery. Physician-defined tumour ROIs on pre-treatment PET were propagated onto the post-treatment PET using rigid and ten deformable registration algorithms. PET images were converted into 256 discrete values. Co-occurrence, run-length, and size zone matrix textures were computed within all ROIs. The relative difference of each texture at different treatment time-points was used to predict the pathologic responders. Their predictive value was assessed using the area under the receiver-operating-characteristic curve (AUC). Propagated ROIs from different algorithms were compared using Dice similarity index (DSI). Contours propagated by the fast-demons, fast-free-form and rigid algorithms did not fully capture the high FDG uptake regions of tumours. Fast-demons propagated ROIs had the least agreement with other contours (DSI = 58%). Moderate to substantial overlap were found in the ROIs propagated by all other algorithms (DSI = 69%-79%). Rigidly propagated ROIs with co-occurrence texture failed to significantly differentiate between responders and non-responders (AUC = 0.58, q-value = 0.33), while the differentiation was significant with other textures (AUC = 0.71-0.73, p < 0.009). Among the deformable algorithms, fast-demons (AUC = 0.68-0.70, q-value < 0.03) and fast-free-form (AUC = 0.69-0.74, q-value < 0.04) were the least predictive. ROIs propagated by all other deformable algorithms with any texture significantly predicted pathologic responders (AUC = 0.72-0.78, q-value < 0.01). Propagated ROIs using deformable registration for all textures can lead to accurate prediction of pathologic response, potentially expediting the temporal texture analysis process. However, fast-demons, fast-free-form, and rigid algorithms should be applied with care due to their inferior performance compared to other algorithms.

Beam's-eye-view imaging during non-coplanar lung SBRT.

PURPOSE: Beam's-eye-view (BEV) imaging with an electronic portal imaging device (EPID) can be performed during lung stereotactic body radiation therapy (SBRT) to monitor the tumor location in real-time. Image quality for each patient and treatment field depends on several factors including the patient anatomy and the gantry and couch angles. The authors investigated the angular dependence of automatic tumor localization during non-coplanar lung SBRT delivery. METHODS: All images were acquired at a frame rate of 12 Hz with an amorphous silicon EPID. A previously validated markerless lung tumor localization algorithm was employed with manual localization as the reference. From ten SBRT patients, 12 987 image frames of 123 image sequences acquired at 48 different gantry-couch rotations were analyzed. δ was defined by the position difference of the automatic and manual localization. RESULTS: Regardless of the couch angle, the best tracking performance was found in image sequences with a gantry angle within 20° of 250° (δ = 1.40 mm). Image sequences acquired with gantry angles of 150°, 210°, and 350° also led to good tracking performances with δ = 1.77-2.00 mm. Overall, the couch angle was not correlated with the tracking results. Among all the gantry-couch combinations, image sequences acquired at (θ = 30°, ϕ = 330°), (θ = 210°, ϕ = 10°), and (θ = 250°, ϕ = 30°) led to the best tracking results with δ = 1.19-1.82 mm. The worst performing combinations were (θ = 90° and 230°, ϕ = 10°) and (θ = 270°, ϕ = 30°) with δ > 3.5 mm. However, 35% (17/48) of the gantry-couch rotations demonstrated substantial variability in tracking performances between patients. For example, the field angle (θ = 70°, ϕ = 10°) was acquired for five patients. While the tracking errors were ≤1.98 mm for three patients, poor performance was found for the other two patients with δ ≥ 2.18 mm, leading to average tracking error of 2.70 mm. Only one image sequence was acquired for all other gantry-couch rotations (δ = 1.18-10.29 mm). CONCLUSIONS: Non-coplanar beams with gantry-couch rotation of (θ = 30°, ϕ = 330°), (θ = 210°, ϕ = 10°), and (θ = 250°, ϕ = 30°) have the highest accuracy for BEV lung tumor localization. Additionally, gantry angles of 150°, 210°, 250°, and 350° also offer good tracking performance. The beam geometries (θ = 90° and 230°, ϕ = 10°) and (θ = 270°, ϕ = 30°) are associated with substantial automatic localization errors. Overall, lung tumor visibility and tracking performance were patient dependent for a substantial number of the gantry-couch angle combinations studied.