CT-based Model Research Articles

Acute Angle of Multilobulated Contours Improves the Risk Classification of Thymomas.

Background: Computed tomography plays an important role in the identification and characterization of thymomas. It has been mainly used during preoperative evaluation for clinical staging. However, the reliable prediction of histological risk types of thymomas based on CT imaging features requires further study. In this study, we developed and validated a nomogram based on CT imaging and included new indices for individualized preoperative prediction of the risk classification of thymomas.Methods: We conducted a retrospective, multicenter study that included 229 patients from two Chinese medical centers. All the patients underwent cross-sectional CT imaging within 2 weeks before surgery. The results of pathological assessments were retrieved from existing reports of the excised lesions. The tumor perimeter that contacted the lung (TPCL) was evaluated and a new quantitative indicator, the acute angle (AA) formed by adjacent lobulations, was measured. Two predictive models of risk classification were created using the least absolute shrinkage and selection operator (LASSO) method in a training cohort for features selection. The model with a smaller Akaike information criterion was then used to create an individualized imaging nomogram, which we evaluated regarding its prediction ability and clinical utility.Results: A new CT imaging-based model incorporating AA was developed and validated, which had improved predictive performance during risk classification of thymomas when compared with a model using traditional imaging predictors. The new imaging nomogram with AA demonstrated its clinical utility by decision curve analysis.Conclusions: Acute angle can improve the performance of a CT-based predictive model during the preoperative risk classification of thymomas and should be considered a new imaging marker for the evaluation and treatment of patients with thymomas. On the contrary, TPCL is not useful as a predictor for the risk classification of thymomas in this study.

608 RADIOMICS MODEL FACILITATES ORGAN-SAVING STRATEGY IN PATIENTS WITH ESOPHAGEAL SQUAMOUS CELL CANCER RECEIVING NEOADJUVANT CHEMORADIOTHERAPY

Abstract After neoadjuvant chemoradiotherapy(NCRT) in locally advanced esophageal squamous cell cancer(ESCC), roughly 40% of the patients may achieve pathologic complete response (pCR) of the primary tumor. Those patients may benefit from organ-saving strategy if the probability of pCR could be correctly identified before esophagectomy. A reliable approach to predict pathological response allows future studies to investigate individualized treatment plans. We aim to establish a CT-based radiomics model to predict tumor response to NCRT. Methods 121 patients with ESCC who underwent NCRT followed by esophagectomy were retrospectively collected. Radiomics features extracted from pre−/post-NCRT CT images were selected by univariate logistic (p < 0.157) and LASSO regression. A radiomics signature(RS) developed with selected features was combined with 4 clinical variables, including percentage of tumor thickness reduction, tumor adventitia type, tumor minimum diameter on post-NCRT esophagogram and age, to construct RS + clinical model with multivariate logistic regression which was internally validated by bootstrapping. Performance and clinical usefulness of RS + clinical model were assessed by receiver operating characteristic(ROC) curves and decision curve analysis, respectively, comparing with the model of clinical variables alone. Results Among the 121 patients, 51 achieved pCR(42%) after NCRT. 16 radiomics features were selected and incorporated into RS. The RS + clinical model has improved prediction performance for pCR compared with the clinical model(corrected area under the ROC curve,0.843 vs. 0.700). At the 60% probability threshold cutoff(i.e., the patient would opt for observation if his probability of pCR was >60%), net 12% surgeries could be avoided by RS + clinical model without an increase in the number of missed residual diseases, equivalent to implementing organ-saving strategy in 29.4% of the 51 true-pCR cases. Conclusion The model built with CT radiomics features and clinical variables shows the potential of predicting pCR after NCRT; it provides significant clinical benefit in identifying qualified patients to receive individualized organ-saving treatment plans.

Radiomic signature of the FOWARC trial predicts pathological response to neoadjuvant treatment in rectal cancer

BackgroundWe aimed to develop a radiomic model based on pre-treatment computed tomography (CT) to predict the pathological complete response (pCR) in patients with rectal cancer after neoadjuvant treatment and tried to integrate our model with magnetic resonance imaging (MRI)-based radiomic signature.MethodsThis was a secondary analysis of the FOWARC randomized controlled trial. Radiomic features were extracted from pre-treatment portal venous-phase contrast-enhanced CT images of 177 patients with rectal cancer. Patients were randomly allocated to the primary and validation cohort. The least absolute shrinkage and selection operator regression was applied to select predictive features to build a radiomic signature for pCR prediction (rad-score). This CT-based rad-score was integrated with clinicopathological variables using gradient boosting machine (GBM) or MRI-based rad-score to construct comprehensive models for pCR prediction. The performance of CT-based model was evaluated and compared by receiver operator characteristic (ROC) curve analysis. The LR (likelihood ratio) test and AIC (Akaike information criterion) were applied to compare CT-based rad-score, MRI-based rad-score and the combined rad-score.ResultsWe developed a CT-based rad-score for pCR prediction and a gradient boosting machine (GBM) model was built after clinicopathological variables were incorporated, with improved AUCs of 0.997 [95% CI 0.990–1.000] and 0.822 [95% CI 0.649–0.995] in the primary and validation cohort, respectively. Moreover, we constructed a combined model of CT- and MRI-based radiomic signatures that achieve better AIC (75.49 vs. 81.34 vs.82.39) than CT-based rad-score (P = 0.005) and MRI-based rad-score (P = 0.003) alone did.ConclusionsThe CT-based radiomic models we constructed may provide a useful and reliable tool to predict pCR after neoadjuvant treatment, identify patients that are appropriate for a 'watch and wait' approach, and thus avoid overtreatment. Moreover, the CT-based radiomic signature may add predictive value to the MRI-based models for clinical decision making.

The Accuracy of Computed Tomography-Based, Three-Dimensional Implant Planning in Robotic-Assisted Total Knee Arthroplasty.

Advanced imaging used in robotic-assisted total knee arthroplasty (TKA), such as computed tomography (CT)-based three-dimensional (3D) planning, may provide an accurate means of implant sizing preoperatively. The purpose of this study was to examine preoperative CT-based implant planning accuracy for robotic-assisted TKA in patients who have (1) varus deformities, (2) valgus deformities, (3) neutral alignment, and (4) retained hardware. A total of 393 patients underwent a robotic-assisted TKA by a single surgeon received preoperative CT scans. The surgeon reviewed the CT-based model preoperatively and recorded the expected size of the components. The final implants used in each case were recorded and compared with the surgeon's preoperative plan. In all groups of patients, the surgeon's CT-based implant plan was within one size of the implant utilized 100% of the time for both the tibiae and femora. Overall, the surgeon was exactly matched in 319 (81%) and 315 (80%) cases for the femoral and tibial components, respectively. For the femoral component, the mean age for patients in whom the original plan was exactly matched was younger than those whose implants were upsized and older than patients those implants were downsized (p = 0.024). Other patient demographics and preoperative knee alignment were not associated with predictive accuracy for femoral or tibial components. Our results demonstrate how preoperative CT-based, 3D planning for robotic-assisted TKA is accurate to within one size of the components in every case (100%), and exactly matched in 80%. The results of this study are important because they demonstrate how CT-based preoperative implant planning for TKA is reliable and accurate across all native knee alignments and other patient-specific factors. In addition, they build on a previous study by the same single surgeon, demonstrating that predictive ability can improve over time. This may be important as we move toward more outpatient surgery with less ability for prostheses inventory at ambulatory sites.

CT-derived Chest Muscle Metrics for Outcome Prediction in Patients with COVID-19.

BackgroundLower muscle mass is a known predictor of unfavorable outcome, but its prognostic impact on COVID-19 patients is unknown.PurposeTo investigate the contribution of CT-derived muscle status in predicting clinical outcomes in COVID-19 patients.Materials and MethodsClinical/laboratory data and outcomes (intensive care unit [ICU] admission and death) were retrospectively retrieved for patients with reverse transcriptase polymerase chain reaction-confirmed COVID-19, who underwent chest CT on admission in four hospitals in Northern Italy from February 21 to April 30, 2020. Extent and type of pulmonary involvement, mediastinal lymphadenopathy, and pleural effusion were assessed. Cross-sectional areas and attenuation of paravertebral muscles were measured on axial CT images at T5 and T12 vertebral level. Multivariable linear and binary logistic regression, including calculation odds ratios (OR) with 95% confidence intervals (CIs), were used to build four models to predict ICU admission and death, tested and compared using receiver operating characteristic curve (ROC) analysis.ResultsA total 552 patients (364 men; median age 65 years, interquartile range 54–75) were included. In a CT-based model, lower-than-median T5 paravertebral muscle area showed the highest ORs for ICU admission (OR 4.8, 95% CI 2.7–8.5; P<.001) and death (OR 2.3, 95% CI 1.0–2.9; P=.027). When clinical variables were included in the model, lower-than-median T5 paravertebral muscle area still showed the highest ORs both for ICU admission (OR 4.3; 95% CI 2.5–7.7; P<.001) and death (OR 2.3, 95% CI 1.3–3.7; P=.001). At ROC analysis, the CT-based model and the model including clinical variables showed the same area under the curve (AUC) for ICU admission prediction (AUC 0.83, P=.380) and were not different in predicting death (AUC 0.86 versus AUC 0.87, respectively, P=.282).ConclusionIn hospitalized patients with COVID-19, lower muscle mass on CT was independently associated with ICU admission and hospital mortality.

Improved PET/MRI accuracy by use of static transmission source in empirically derived hardware attenuation correction

BackgroundAccurate quantification of radioactivity, measured by an integrated positron emission tomography (PET) and magnetic resonance imaging (MRI) system, is still a challenge. One aspect of such a challenge is to correct for the hardware attenuation, such as the patient table and radio frequency (RF) resonators. For PET/MRI systems, computed tomography (CT) is commonly used to produce hardware attenuation correction (AC) maps, by converting Hounsfield units (HU) to a linear attenuation coefficients (LAC) map at the PET energy level 511 keV, using a bilinear model. The model does not address beam hardening, nor higher density materials, which can lead to inaccurate corrections.PurposeIn this study, we introduce a transmission-based (TX-based) AC technique with a static Germanium-68 (Ge-68) transmission source to generate hardware AC maps using the PET/MRI system itself, without the need for PET or medical CT scanners. The AC TX-based maps were generated for a homogeneous cylinder, made of acrylic as a validator. The technique thereafter was applied to the patient table and posterior part of an RF-phased array used in cardiovascular PET/MRI imaging. The proposed TX-based, and the CT-based, hardware maps were used in reconstructing PET images of one cardiac patient, and the results were analysed and compared.ResultsThe LAC derived by the TX-based method for the acrylic cylinder is estimated to be 0.10851 ± 0.00380 cm−1 compared to the 0.10698 ± 0.00321 cm−1 theoretical value reported in the literature. The PET photon counts were reduced by 8.7 ± 1.1% with the patient table, at the region used in cardiac scans, while the CT-based map, used for correction, over-estimated counts by 4.3 ± 1.3%. Reconstructed in vivo images using TX-based AC hardware maps have shown 4.1 ± 0.9% mean difference compared to those reconstructed images using CT-based AC.ConclusionsThe LAC of the acrylic cylinder measurements using the TX-based technique was in agreement with those in the literature confirming the validity of the technique. The over-estimation of photon counts caused by the CT-based model used for the patient table was improved by the TX-based technique. Therefore, TX-based AC of hardware using the PET/MRI system itself is possible and can produce more accurate images when compared to the CT-based hardware AC in cardiac PET images.

Clinical utility of a computed tomography-based receiver operating characteristic curve model for the diagnosis of COVID-19.

The outbreak of COVID-19 poses a major and urgent threat to global public health. CT findings associated with COVID-19 pneumonia from initial diagnosis until patient recovery. This study aimed to retrospectively analyze abnormal lung changes following initial computed tomography (CT) among patients with coronavirus disease 2019 (COVID-19) in Yunnan, and to evaluate the effectiveness of a chest CT-based model for the diagnosis of COVID-19. One hundred and nine patients with COVID-19 pneumonia confirmed with the positive new coronavirus nucleic acid antibody who exhibited abnormal findings on initial CT were retrospectively analyzed. Thereafter, changes in the number, distribution, shape, and density of the lesions were observed. Further, the epidemiological, clinical, and CT imaging findings (+/-) were correlated. Following univariate and multivariate logistic regression analysis, receiver operating characteristic (ROC) curves were generated for significant factors, and models were established to evaluate the diagnostic ability of CT for COVID-19. Our results showed significant differences between patients with COVID-19 in epidemiological history (first, second, and third generation), clinical type (moderate, severe, and critical), and abnormal CT imaging characteristics (+/-) (P<0.05). Moreover, significant differences in abnormal CT imaging characteristics, including region, extent, and focus, were observed between the first generation and the other generations (P<0.05). For the diagnosis of COVID-19, the areas under the ROC curves for logistic regression models 1, 2, and 3 were 0.8016 (95% CI: 0.6759-0.9274), 0.9132 (95% CI: 0.8571-0.9693), and 0.9758 (95% CI: 0.9466-1), respectively. The ROC curve regression model based on chest CT signs displayed a high diagnostic value for COVID-19.

Visceral adipose tissue area predicts intensive care unit admission in COVID-19 patients

We retrospectively investigated, in 62 consecutive hospitalised COVID-19 patients (aged 70 ± 14 years, 40 males), the prognostic value of CT-derived subcutaneous adipose tissue and visceral adipose tissue (VAT) metrics, testing them in four predictive models for admission to intensive care unit (ICU), with and without pre-existing comorbidities. Multivariate logistic regression identified VAT score as the best ICU admission predictor (odds ratios 4.307–12.842). A non-relevant contribution of comorbidities at receiver operating characteristic analysis (area under the curve 0.821 for the CT-based model, 0.834 for the one including comorbidities) highlights the potential one-stop-shop prognostic role of CT-derived lung and adipose tissue metrics.

Survival Prediction in Gallbladder Cancer Using CT Based Machine Learning.

ObjectiveTo establish a classifier for accurately predicting the overall survival of gallbladder cancer (GBC) patients by analyzing pre-treatment CT images using machine learning technology.MethodsThis retrospective study included 141 patients with pathologically confirmed GBC. After obtaining the pre-treatment CT images, manual segmentation of the tumor lesion was performed and LIFEx package was used to extract the tumor signature. Next, LASSO and Random Forest methods were used to optimize and model. Finally, the clinical information was combined to accurately predict the survival outcomes of GBC patients.ResultsFifteen CT features were selected through LASSO and random forest. On the basis of relative importance GLZLM-HGZE, GLCM-homogeneity and NGLDM-coarseness were included in the final model. The hazard ratio of the CT-based model was 1.462(95% CI: 1.014–2.107). According to the median of risk score, all patients were divided into high and low risk groups, and survival analysis showed that high-risk groups had a poor survival outcome (P = 0.012). After inclusion of clinical factors, we used multivariate COX to classify patients with GBC. The AUC values in the test set and validation set for 3 years reached 0.79 and 0.73, respectively.ConclusionGBC survival outcomes could be predicted by radiomics based on LASSO and Random Forest.

PH-0286: Development of CT-based cardiac model with substructure for dosimetry in late effects studies

PH-0286: Development of CT-based cardiac model with substructure for dosimetry in late effects studies

A Pre-Treatment PET And CT-Derived Model for Progression-Free Survival in Early Stage Non-Small Cell Lung Cancer

The maximum value of the standardized uptake value (SUV) and tumor diameter have been associated with progression after stereotactic body radiotherapy (SBRT) for early stage non-small cell lung cancer (NSCLC). In this work, we explored whether other tumor parameters based on pre-SBRT PETs and CTs predict progression-free survival (PFS) in this patient population. 412 patients treated between 2006 and 2017 were included. Patients had to have PETs and CTs available within three months prior to SBRT start. The median prescription dose was 50Gy in 5 fractions. The planning-CT gross tumor volumes (GTVs) were propagated onto the pre-treatment PETs and CTs using b-spline deformable image registration. PET intensity features (90th percentile, entropy, maximum, mean, peak, robust mean absolute deviation, SD, valley) and CT shape features (compactness, diameter, elliptic axes, elongation, flatness, number of lobules/peaks/spicules, sphericity, surface area, surface to volume ratio, volume) were extracted. Data were split into training and hold-out validation subsets (n = 283, 123; 70%, 30%). In the training subset, the imaging features and six patient characteristics (age, gender, histology, performance status, prior surgery, tumor location) were tested for association with PFS using Cox Proportional Hazards regression with re-sampling (bootstrapping with 1000 samples). Significance was denoted at p≤0.0019 (corrected for 26 tests). A bootstrapped forward-stepwise multivariate analysis was undertaken including only non-strongly correlated predictors (Spearman’s rank, |Rs|<0.70). The most frequently selected model was explored in the validation subset in which model performance was assessed using the c-index and the prediction-stratified high and low risk tertiles (HR, LR) of the observed PFS were compared. Nineteen of the 20 identified candidate predictors were either PET or CT features (p-value range: 3E-9, 1.2E-3). The intra-imaging modality correlation between features was strong (median |Rs|: PET: 0.93; CT: 0.76) and only four features were passed on to multivariate analysis: PET entropy, CT number of peaks, CT major axis, and gender. The most frequently selected model included PET entropy and CT number of peaks; the c-index in the validation subset was 0.77 and the prediction-stratified survival indicated a clear separation between the observed HR and LR: e.g. a PFS of 60% was observed at 12 months in HR vs. 22 months in LR. This PET and CT-based model identified the SUV distribution randomness (entropy) and spiculated tumor pattern on CTs as the most important features in predicting PFS in early stage NSCLC. The associated performance on the hold-out validation subset was good and its use has the potential to further improve the prediction of response to SBRT for this patient population. This model will be used to identify high-risk patients based on the predicted PFS in an upcoming phase II study on adjuvant immunotherapy.

Computed tomography-based deep-learning prediction of neoadjuvant chemoradiotherapy treatment response in esophageal squamous cell carcinoma

BackgroundDeep learning is promising to predict treatment response. We aimed to evaluate and validate the predictive performance of the CT-based model using deep learning features for predicting pathologic complete response to neoadjuvant chemoradiotherapy (nCRT) in esophageal squamous cell carcinoma (ESCC). Materials and methodsPatients were retrospectively enrolled between April 2007 and December 2018 from two institutions. We extracted deep learning features of six pre-trained convolutional neural networks, respectively, from pretreatment CT images in the training cohort (n = 161). Support vector machine was adopted as the classifier. Validation was performed in an external testing cohort (n = 70). We assessed the performance using the area under the receiver operating characteristics curve (AUC) and selected an optimal model, which was compared with a radiomics model developed from the training cohort. A clinical model consisting of clinical factors only was also built for baseline comparison. We further conducted a radiogenomics analysis using gene expression profiles to reveal underlying biology associated with radiological prediction. ResultsThe optimal model with features extracted from ResNet50 achieved an AUC and accuracy of 0.805 (95% CI, 0.696–0.913) and 77.1% (65.6%-86.3%) in the testing cohort, compared with 0.725 (0.605–0.846)) and 67.1% (54.9%-77.9%) for the radiomics model. All the radiological models showed better predictive performance than the clinical model. Radiogenomics analysis suggested a potential association mainly with WNT signaling pathway and tumor microenvironment. ConclusionsThe novel and noninvasive deep learning approach could provide efficient and accurate prediction of treatment response to nCRT in ESCC, and benefit clinical decision making of therapeutic strategy.

Geometric and Volumetric Relationship Between Human Lumbar Vertebra and CT-based Models

Crucial to the process of three-dimensional (3D) printing is the knowledge of how the actual structure or organ relates dimensionally to its corresponding medical image. This study will examine the differences between human lumbar vertebrae, 3D scans of these bones, 3D models based on computed tomographic (CT) scans, and 3D-printed models. CT scans were obtained for six human lumbar spines. The bones were cleaned, and 3D scanned. 3D mesh models were created from the CT data, and then 3D printed. Four models were analyzed: anatomic bones, 3D-scanned models, CT-models, and 3D-printed models. Manual measurements were performed for all model types, and segmentation metric comparisons were performed comparing the 3D-scanned models to the CT-models. There was no statistical difference between manual measurements when comparing each parameter of all model types, except for vertebral width (p = 0.044). There was no statistical difference when comparing the average of all measurements between all model types (p = 0.247). The mean Hausdorff distance was 0.99 mm (SD 0.55 mm) when comparing 3D-scanned model to CT-model. The mean Dice coefficient was 0.90 (SD 0.07) when comparing 3D-scanned model to CT-model. The mean volume for 3D-scanned model and CT-model were 41.6 ml and 45.9 ml (p < 0.001), respectively. This study clarifies the geometric and volumetric relationship between human lumbar vertebra and CT-based vertebral models. Segmentation metrics reveal a 1 mm difference between examined bones (using the 3D-scanned bone as a surrogate), and the CT measurements. This is confirmed by a volumetric difference of 4.3 ml, between the larger CT-based model and the smaller bone.

A CT-based radiomics model to detect prostate cancer lymph node metastases in PSMA radioguided surgery patients

PurposeIn recurrent prostate carcinoma, determination of the site of recurrence is crucial to guide personalized therapy. In contrast to prostate-specific membrane antigen (PSMA)–positron emission tomography (PET) imaging, computed tomography (CT) has only limited capacity to detect lymph node metastases (LNM). We sought to develop a CT-based radiomic model to predict LNM status using a PSMA radioguided surgery (RGS) cohort with histological confirmation of all suspected lymph nodes (LNs).MethodsEighty patients that received RGS for resection of PSMA PET/CT-positive LNMs were analyzed. Forty-seven patients (87 LNs) that received inhouse imaging were used as training cohort. Thirty-three patients (62 LNs) that received external imaging were used as testing cohort. As gold standard, histological confirmation was available for all LNs. After preprocessing, 156 radiomic features analyzing texture, shape, intensity, and local binary patterns (LBP) were extracted. The least absolute shrinkage and selection operator (radiomic models) and logistic regression (conventional parameters) were used for modeling.ResultsTexture and shape features were largely correlated to LN volume. A combined radiomic model achieved the best predictive performance with a testing-AUC of 0.95. LBP features showed the highest contribution to model performance. This model significantly outperformed all conventional CT parameters including LN short diameter (AUC 0.84), LN volume (AUC 0.80), and an expert rating (AUC 0.67). In lymph node–specific decision curve analysis, there was a clinical net benefit above LN short diameter.ConclusionThe best radiomic model outperformed conventional measures for detection of LNM demonstrating an incremental value of radiomic features.

Development and validation of a CT-based radiomic model combined with margin-related radiomic features to distinguish precancerous lesions from early-stage lung adenocarcinoma.

e21044 Background: The purpose of this study was to investigate whether the combined radiomic model based on tumor-associated and margin-related (5mm) radiomic features can effectively improve prediction performance of distinguishing precancerous lesions from early stage lung adenocarcinoma. Methods: 264 patients underwent preoperative chest CT in Guangdong Provincial People’s hospital from March 1, 2015 to December 31,2019 were sorted by three cohorts. All lesions were pathologically confirmed as precancerous lesions or Stage I lung adenocarcinoma and a total of 861 analyzable radiomic features were extracted from two segmented lesions including pulmonary lesions and margins, using PyRadiomics by two senior radiologists. In training cohort, 145 patients (70%) are selected randomly from the single-nodular patients (N = 207). As for the validation cohorts, the models were validated using the resting 62 patients from single-nodular cohort and multi-nodular cohort (n = 57) respectively. Least Absolute Shrinkage and Selector Operation and Support Vector Machine-Recursive Feature Elimination were used for feature selection. ROC analysis and AUC were used to evaluate the performance of three models which were developed by multiple logistic regression on distinguishing the precancerous lesions from early stage lung adenocarcinoma. Results: Selected features from pulmonary lesions and pericarcinous tissue were developed into two independent radiomic models and a combined model. Margin-related radiomic model performs well in three validation cohorts. The AUC Brock of single-nodular cohort in training cohort was 0.912 (95% CI: 0.876-0.948), while in single-nodular validation cohort was 0.93 (95% CI: 0.862-0.966). Multi-nodular validation cohort in this model shows an AUC of 0.891 (95% CI = 0.824–0.943). Comparing combined model and tumor-associated radiomic model, it is found that the AUC of combined model was improved from 0.865 (95% CI: 0.767-0.963) to 0.94 (95% CI: 0.767-0.963) for single-nodular validation cohort. Respectively, this combined model also performs well in multi-nodular validation cohort. Conclusions: This study demonstrated the potential of margin-related radiomic features based on preoperative CT scans to distinguish precancerous lesions from early stage lung adenocarcinoma. The constructed radiomic model provided an easy-to-use, preoperative tool for surgeons to develop accurate therapeutic strategies for multi-nodular patients.

Prediction of esophageal fistula from esophageal cancer CT images using multi-view multi-scale attentional convolutional neural network (MM-Atten-CNN).

4553 Background: We aimed to propose a risk model based on MM-Atten-CNN for predicting esophageal fistula in patients with esophageal cancer (EC) from computerized tomography (CT) -based radiomics. Methods: EC patients who didn’t received esophageal surgery between July 2014 and August 2019 were collected. Of these, 186 patients (cases) who developed esophageal fistula were enrolled and compared with 372 controls (1:2 matched with the diagnosis time of EC, sex, marriage, and race). All 558 patients were divided into training set (n = 390) and validation set (n = 168) randomly. The MM-Atten-CNN risk model was trained over 2D slices from nine views of planes, where there were three patches of contextual CT, segmented tumor and neighbouring information in each view. In the training set (130 cases and 260 controls), data augmentation was performed including pixel shifting [-10, -5, +5, +10] and rotation [-10, +10]. In total, there were (130+260) *16*2 = 12480 subjects used for training. Finally, the risk model was validated in the validation set (56 cases and 112 controls) and measured by accuracy (acc), sensitivity (sen), and specificity (spe). Results: The developed risk model achieved (acc, sen, spe) of (0.839, 0.807, 0.926), which were more predictive for the occurrence of esophageal fistula when compared to CNN models using single coronal view (acc 0.763, sen 0.581, spe 0.837), multi-view 2D contextual CT slices (acc 0.779, sen 0.656, spe 0.896), and 3D CNN using contextual CT volumes (acc 0.781, sen 0.689, spe 0.852). Conclusions: MM-Atten-CNN CT-based model improved the performance of esophageal fistula risk prediction, which has the potential to assist individualized stratification and treatment planning in EC patients.

Fully automated dose prediction using generative adversarial networks in prostate cancer patients.

PurposeAlthough dose prediction for intensity modulated radiation therapy (IMRT) has been accomplished by a deep learning approach, delineation of some structures is needed for the prediction. We sought to develop a fully automated dose-generation framework for IMRT of prostate cancer by entering the patient CT datasets without the contour information into a generative adversarial network (GAN) and to compare its prediction performance to a conventional prediction model trained from patient contours.MethodsWe propose a synthetic approach to translate patient CT datasets into a dose distribution for IMRT. The framework requires only paired-images, i.e., patient CT images and corresponding RT-doses. The model was trained from 81 IMRT plans of prostate cancer patients, and then produced the dose distribution for 9 test cases. To compare its prediction performance to that of another trained model, we created a model trained from structure images. Dosimetric parameters for the planning target volume (PTV) and organs at risk (OARs) were calculated from the generated and original dose distributions, and mean differences of dosimetric parameters were compared between the CT-based model and the structure-based model.ResultsThe mean differences of all dosimetric parameters except for D98% and D95% for PTV were within approximately 2% and 3% of the prescription dose for OARs in the CT-based model, while the differences in the structure-based model were within approximately 1% for PTV and approximately 2% for OARs, with a mean prediction time of 5 seconds per patient.ConclusionsAccurate and rapid dose prediction was achieved by the learning of patient CT datasets by a GAN-based framework. The CT-based dose prediction could reduce the time required for both the iterative optimization process and the structure contouring, allowing physicians and dosimetrists to focus their expertise on more challenging cases.

MP55-06 OUTCOME OF NON-RESPONDERS TO NEOADJUVANT IMMUNOTHERAPY COMPARED TO CISPLATIN-BASED CHEMOTHERAPY BEFORE RADICAL CYSTECTOMY IN MUSCLE-INVASIVE BLADDER CANCER

INTRODUCTION AND OBJECTIVE: Patients (pts) with ypT3-4 or ypN+ response after neoadjuvant chemotherapy (CT) have a dismal prognosis regardless the use of adjuvant therapies. It is possible that similar pathologic responses after neoadjuvant immunotherapy (IO) portend different prognosis. Here, we compared the relapse-free survival (RFS) after RC in ypT3-4 or yN+ pts after IO vs CT METHODS: Between 2014 and 2019, we identified all patients with cT2-4N0 MIBC treated with neoadjuvant CT or IO (pembrolizumab), followed by RC, at our tertiary care institution. For this study purpose, only ypT3-4 or ypN+ pts were considered. Kaplan-Meier analyses examined the RFS according to neoadjuvant therapy group (CT VS IO). Cox regression analysis examined the risk of relapse after adjusting for pT, pN, age at surgery, gender and use of adjuvant CT. Finally, actual rate of 1-year RFS after RC was plotted against pT, pN, and adjuvant CT-based risk of recurrence for pts treated with neoadjuvant CT or IO RESULTS: We retrospectively identified 37 and 31 ypT3-4 or ypN+ MIBC pts. Median follow-up was 12months (interquartile range [IQR] 6-17months). No baseline differences emerged between neoadjuvant treatment groups in terms of age, gender, ypT stage, ypN stage, and use of adjuvant CT (all p>0.1). At Kaplan-Meier and multivariable Cox-regression analyses, non-responder pts treated with IO (pembrolizumab) showed higher 1-year RFS (79% vs. 47%, adjusted HR: 0.32, 95% CI: 0.11-0.93, p=0.03) compared with patients treated with neoadjuvant CT. The interaction between the use of neoadjuvant therapies (IO vs. CHT) and the risk of 1-year recurrence derived from ypT, ypN, and use of adjuvant CT-based model was statistically significant (p=0.0014). Here, non-responder neoadjuvant IO-treated patients presented a clinically significant lower rate of 1-year recurrence for predicted probabilities of recurrence higher than 20%. CONCLUSIONS: The paradigm that non-responder ypT3-4 or ypN+ MIBCa patients present adverse oncologic outcomes was confirmed after standard neoadjuvant chemotherapy, but not after neoadjuvant pembrolizumab. Neoadjuvant immunotherapy still shows an oncologic benefit even in patients with apparently unresponsive disease.Source of Funding: none

A computed tomography-based nomogram for preoperative prediction of synchronous peritoneal carcinomatosis in colorectal cancer.

14 Background: Peritoneal carcinomatosis (PC) of colorectal cancer remain poor and is considered to be the terminal stage. Early detection of PC can bring survival benefit for chances of complete cytoreductive surgery. In this study, we have developed and validated a CT-based nomogram for preoperative prediction of synchronous PC in CRC. Methods: The CT-based nomogram optimized for image classification of PC was developed and validated in a large cohort. The specific features of synchronous PC were extracted from preoperative CT arterial stage scans by radiologists. A CT-based model was constructed by Boruta algorithm and multivariate logistic regression in the training set and validated in the test set. A CT-based nomogram was then built to predict the risk value of PC. The performance and efficacy of this nomogram to discriminate PC were assessed. Results: A total of 170 eligible CRC patients with synchronous PC were randomly divided into a training set (n = 92) and a validation set (n = 78). After processing and partition, five items of PC-related features including tumor location, cT stage, distant metastasis, thickened greater omentum and pelvic nodules were enrolled into this CT-based model. In the training set, the ROC shows AUC of CT-based model was 0.929(95% CI: 0.764-0.946). The sensitivity was 73.2% and the specificity was 98.3%. The positive predictive value (PPV) was 82.26%, while the negative predictive value (NPV) was 100%. In the test set, AUC by the ROC was 0.855(95% CI: 0.764-0.946). The sensitivity of CT-based model was 72.7% and the specificity was 93.3%. The PPV was 82.35%, while the NPV was 88.89%. These results show high predicted accuracy of CT-based model in diagnosis of synchronous PC both in the training set and test set. A nomogram based on these five preoperative predictors of PC was created. This CT-based nomogram showed good diagnostic performance by calibration curve. The decision curves analysis revealed high benefit of this CT-based nomogram in clinical decision. Conclusions: The CT-based nomogram has shown great potential in the detection and diagnosis of synchronous peritoneal metastasis in colorectal cancer.

Computed Tomography-Based Modeling of Water Vapor-Induced Changes in Permittivity During Microwave Ablation.

Measurements of tissue permittivity with small open-ended coaxial probes during microwave tissue heating have been plagued by high variability as tissue water becomes vaporized. Analysis of such variability has been hampered by a lack of direct visualization of the measurement volume. The objective of this study was to determine if X-ray computed tomography (CT) could be used to visualize the measurement volume and then predict dielectric permittivity based on the visualized tissue composition. CT attenuation at 120kVp was measured at the end of an open-ended coaxial probe during microwave ablation of ex vivo liver (2.45GHz at 50 W delivered for 5minutes). Tissue composition was estimated from attenuation maps and used to predict dielectric properties based on established mixture equations. Predicted permittivity was then compared to measured values using error metrics and linear regression. There was a good agreement between measured and modeled permittivity during 5-minute ablations at 2.45GHz (r = 0.94, p < .001). Normalized root mean squared errors were below 17% in permittivity modeling at 2.45GHz. CT-based model predictions of dielectric properties are feasible in ex vivo liver. The models may facilitate real-time imaging-based permittivity mapping.