Search

Background and purpose: Accurate instance segmentation of teeth in panoramic dental X-rays is a challenging task due to variations in tooth morphology and overlapping regions. In this study, we propose a new algorithm, for instance, segmentation of the different teeth in panoramic dental X-rays. Methods: An instance segmentation model was trained using the architecture of a Mask Region-based Convolutional Neural Network (Mask-RCNN). The data for the training, validation, and testing were taken from the Tuft dental database (1000 panoramic dental radiographs). The number of the predicted label was 52 (20 deciduous and 32 permanent). The size of the training, validation, and test sets were 760, 190, and 70 images, respectively, and the split was performed randomly. The model was trained for 300 epochs, using a batch size of 10, a base learning rate of 0.001, and a warm-up multistep learning rate scheduler (gamma = 0.1). Data augmentation was performed by changing the brightness, contrast, crop, and image size. The percentage of correctly detected teeth and Dice in the test set were used as the quality metrics for the model. Results: In the test set, the percentage of correctly classified teeth was 98.4%, while the Dice score was 0.87. For both the left mandibular central and lateral incisor permanent teeth, the Dice index result was 0.91 and the accuracy was 100%. For the permanent teeth right mandibular first molar, mandibular second molar, and third molar, the Dice indexes were 0.92, 0.93, and 0.78, respectively, with an accuracy of 100% for all three different teeth. For deciduous teeth, the Dice indexes for the right mandibular lateral incisor, right mandibular canine, and right mandibular first molar were 0.89, 0.91, and 0.85, respectively, with an accuracy of 100%. Conclusions: A successful instance segmentation model for teeth identification in panoramic dental X-ray was developed and validated. This model may help speed up and automate tasks like teeth counting and identifying specific missing teeth, improving the current clinical practice.

Purpose To evaluate stability and machine learning-based classification performance of radiomic features of spine bone tumors using diffusion- and T2-weighted magnetic resonance imaging (MRI). Material and methods This retrospective study included 101 patients with histology-proven spine bone tumor (22 benign; 38 primary malignant; 41 metastatic). All tumor volumes were manually segmented on morphologic T2-weighted sequences. The same region of interest (ROI) was used to perform radiomic analysis on ADC map. A total of 1702 radiomic features was considered. Feature stability was assessed through small geometrical transformations of the ROIs mimicking multiple manual delineations. Intraclass correlation coefficient (ICC) quantified feature stability. Feature selection consisted of stability-based (ICC > 0.75) and significance-based selections (ranking features by decreasing Mann–Whitney p-value). Class balancing was performed to oversample the minority (i.e., benign) class. Selected features were used to train and test a support vector machine (SVM) to discriminate benign from malignant spine tumors using tenfold cross-validation. Results A total of 76.4% radiomic features were stable. The quality metrics for the SVM were evaluated as a function of the number of selected features. The radiomic model with the best performance and the lowest number of features for classifying tumor types included 8 features. The metrics were 78% sensitivity, 68% specificity, 76% accuracy and AUC 0.78. Conclusion SVM classifiers based on radiomic features extracted from T2- and diffusion-weighted imaging with ADC map are promising for classification of spine bone tumors. Radiomic features of spine bone tumors show good reproducibility rates.

Wall shear stress (WSS) is involved in coronary artery plaque pathological mechanisms and modulation of gene expression. This study aims to provide a comprehensive haemodynamic and biological description of unstable (intact-fibrous-cap, IFC, and ruptured-fibrous-cap, RFC) and stable (chronic coronary syndrome, CCS) plaques and investigate any correlation between WSS and molecular pathways.We enrolled 24 CCS and 25 Non-ST Elevation Myocardial Infarction-ACS patients with IFC (n = 11) and RFC (n = 14) culprit lesions according to optical coherence tomography analysis. A real-time PCR primer array was performed on peripheral blood mononuclear cells for 17 different molecules whose expression is linked to WSS. Computational fluid dynamics simulations were performed in high-fidelity 3D-coronary artery anatomical models for three patients per group. A total of nine genes were significantly overexpressed in the unstable patients as compared to CCS patients, with no differences between IFC and RFC groups (GPX1, MMP1, MMP9, NOS3, PLA2G7, PI16, SOD1, TIMP1, and TFRC) while four displayed different levels between IFC and RFC groups (TNFα, ADAMTS13, EDN1, and LGALS8). A significantly higher WSS was observed in the RFC group (p 0.001) compared to the two other groups. A significant correlation was observed between TNFα (p 0.001), EDN1 (p = 0.036), and MMP9 (p = 0.005) and WSS values in the RFC group.Our data demonstrate that IFC and RFC plaques are subject to different WSS conditions and gene expressions, suggesting that WSS profiling may play an essential role in the plaque instability characterization with relevant diagnostic and therapeutic implications in the era of precision medicine.

BACKGROUND AND PURPOSE: Prognosis in locally advanced head and neck cancer (HNC) is currently based on TNM staging system and tumor subsite. However, quantitative imaging features (i.e., radiomic features) from magnetic resonance imaging (MRI) may provide additional prognostic info. The aim of this work is to develop and validate an MRI-based prognostic radiomic signature for locally advanced HNC.MATERIALS AND METHODS: Radiomic features were extracted from T1- and T2-weighted MRI (T1w and T2w) using the segmentation of the primary tumor as mask. In total 1072 features (536 per image type) were extracted for each tumor. A retrospective multi-centric dataset (n = 285) was used for features selection and model training. The selected features were used to fit a Cox proportional hazard regression model for overall survival (OS) that outputs the radiomic signature. The signature was then validated on a prospective multi-centric dataset (n = 234). Prognostic performance for OS and disease-free survival (DFS) was evaluated using C-index. Additional prognostic value of the radiomic signature was explored.RESULTS: The radiomic signature had C-index = 0.64 for OS and C-index = 0.60 for DFS in the validation set. The addition of the radiomic signature to other clinical features (TNM staging and tumor subsite) increased prognostic ability for both OS (HPV- C-index 0.63 to 0.65; HPV+ C-index 0.75 to 0.80) and DFS (HPV- C-index 0.58 to 0.61; HPV+ C-index 0.64 to 0.65).CONCLUSION: An MRI-based prognostic radiomic signature was developed and prospectively validated. Such signature can successfully integrate clinical factors in both HPV+ and HPV- tumors.

Aim of this study is to assess the repeatability of radiomic features on magnetic resonance images (MRI) and their stability to variations in time of repetition (TR), time of echo (TE), slice thickness (ST), and pixel spacing (PS) using vegetable phantoms.The organic phantom was realized using two cucumbers placed inside a cylindrical container, and the analysis was performed using T1-weighted (T1w), T2-weighted (T2w), and diffusion-weighted images. One dataset was used to test the repeatability of the radiomic features, whereas other four datasets were used to test the sensitivity of the different MRI sequences to image acquisition parameters (TR, TE, ST, and PS). Four regions of interest (ROIs) were segmented: two for the central part of each cucumber and two for the external parts. Radiomic features were extracted from each ROI using Pyradiomics. To assess the effect of preprocessing on the reduction of variability, features were extracted both before and after the preprocessing. The coefficient of variation (CV) and intra-class correlation coefficient (ICC) were used to evaluate variability.The use of intensity standardization increased the stability for the first-order statistics features. Shape and size features were always stable for all the analyses. Textural features were particularly sensitive to changes in ST and PS, although some increase in stability could be obtained by voxel size resampling. When images underwent image preprocessing, the number of stable features (ICC 0.75 and mean absolute CV 0.3) was 33 for apparent diffusion coefficient (ADC), 52 for T1w, and 73 for T2w.The most critical source of variability is related to changes in voxel size (either caused by changes in ST or PS). Preprocessing increases features stability to both test-retest and variation of the image acquisition parameters for all the types of analyzed MRI (T1w, T2w, and ADC), except for ST.

Purpose To study a robust and reproducible procedure to investigate a relation between focal brain radiotherapy (RT) low doses, neurocognitive impairment and late White Matter and Gray Matter alterations, as shown by Diffusion Tensor Imaging (DTI), in children. Methods and materials Forty-five patients (23 males and 22 females, median age at RT 6.2 years, median age at evaluations 11.1 years) who had received focal RT for brain tumors were recruited for DTI exams and neurocognitive tests. Patients’ brains were parceled in 116 regions of interest (ROIs) using an available segmented atlas. After the development of an ad hoc, home-made, multimodal and highly deformable registration framework, we collected mean RT doses and DTI metrics values for each ROI. The pattern of association between cognitive scores or domains and dose or DTI values was assessed in each ROI through both considering and excluding ROIs with mean doses higher than 75% of the prescription. Subsequently, a preliminary threshold value of dose discriminating patients with and without neurocognitive impairment was selected for the most relevant associations. Results The workflow allowed us to identify 10 ROIs where RT dose and DTI metrics were significantly associated with cognitive tests results (p Conclusion This analysis, despite being conducted on a retrospective cohort of children, shows that the identification of critical brain structures and respective radiation dose thresholds is achievable by combining, with appropriate methodological tools, the large amount of data arising from different sources. This supported the design of a prospective study to gain stronger evidence.

Objectives: To identify and validate baseline magnetic resonance imaging (b-MRI) radiomic features (RFs) as predictors of disease outcomes in effectively cured head and neck squamous cell carcinoma (HNSCC) patients. Materials and methods: Training set (TS) and validation set (VS) were retrieved from preexisting datasets (HETeCo and BD2Decide trials, respectively). Only patients with both pre- and post-contrast enhancement T1 and T2-weighted b-MRI and at least 2years of follow-up (FUP) were selected. The combination of the best extracted RFs was used to classify low risk (LR) vs. high risk (HR) of disease recurrence. Sensitivity, specificity, and area under the curve (AUC) of the radiomic model were computed on both TS and VS. Overall survival (OS) and 5-year disease-free survival (DFS) Kaplan–Meier (KM) curves were compared for LR vs. HR. The radiomic-based risk class was used in a multivariate Cox model, including well-established clinical prognostic factors (TNM, sub-site, human papillomavirus [HPV]). Results: In total, 57 patients of TS and 137 of VS were included. Three RFs were selected for the signature. Sensitivity of recurrence risk classifier was 0.82 and 0.77, specificity 0.78 and 0.81, AUC 0.83 and 0.78 for TS and VS, respectively. VS KM curves for LR vs. HR groups significantly differed both for 5-year DFS (p

Background The use of Optical Coherence Tomography (OCT) in acute coronary syndromes (ACS) allows recognizing ruptured fibrous cap (RFC) and intact fibrous cap (IFC) culprit lesions. The biological differences between them, as recently pointed out in translation studies, highlight different mechanisms for a similar clinical manifestation that might deserve different therapeutic approaches. The relationship between endothelial wall shear stress (WSS) and ACS has been demonstrated, however the differences in WSS features between RFC and IFC have not been elucidated. Purpose The aim of this study is to provide a fluid-dynamic and biological description of unstable and stable (SA) plaques, according to OCT analysis. Methods We enrolled 10 SA and 20 Non-ST Elevation Myocardial Infarction (NSTEMI)-ACS patients, with IFC (n=10) and RFC (n=10) culprit lesions according to OCT analysis. We performed Real-time PCR primer array on pooled Peripheral Blood Mononuclear Cell (PBMC) for 30 different molecules whose expression is strictly dependent on WSS. High-fidelity 3D-coronary artery models were created for 3 patients per group, applying previously validated methodologies. Results Among the groups we found a broad difference in molecular expression (Fig. 1A), with RFC displaying higher levels of molecules involved in vasoconstriction/dilatation (EDN1, NOS3), cellular adhesion (ICAM1), and peptidase inhibition (PI16). A significantly higher WSS was observed in RFC group (p Conclusions Our data demonstrated that IFC and RFC unstable plaques are associated with different WSS conditions, alongside with the expression of different molecular patterns specifically related to altered WSS. In the era of precision medicine these findings may have relevant therapeutical implications. Funding Acknowledgement Type of funding source: None

Simple Summary The prognostic performance of traditional methodologies in advanced nasopharyngeal carcinoma does not allow to successfully stratify patients. Previous studies showed that MRI-radiomics has been used to give additional information to improve the prognosis for this type of pathology in patients from endemic areas (Asia). The purpose of this study was to use MRI-radiomics to develop prognostic models for overall survival in patients from non-endemic areas (Europe or United States). In particular, T1-weighted and T2-weighted MRI were used for the purpose. Radiomic features from those images allowed to successfully train a prognostic signature that improved the prognostic performance of models based on clinical variables alone for different clinical endpoints (overall survival, disease-free survival and loco-regional recurrence-free survival). These results suggest how MRI-radiomics is a useful additional tool for prognosis in nasopharyngeal cancer. Abstract Advanced stage nasopharyngeal cancer (NPC) shows highly variable treatment outcomes, suggesting the need for independent prognostic factors. This study aims at developing a magnetic resonance imaging (MRI)-based radiomic signature as a prognostic marker for different clinical endpoints in NPC patients from non-endemic areas. A total 136 patients with advanced NPC and available MRI imaging (T1-weighted and T2-weighted) were selected. For each patient, 2144 radiomic features were extracted from the main tumor and largest lymph node. A multivariate Cox regression model was trained on a subset of features to obtain a radiomic signature for overall survival (OS), which was also applied for the prognosis of other clinical endpoints. Validation was performed using 10-fold cross-validation. The added prognostic value of the radiomic features to clinical features and volume was also evaluated. The radiomics-based signature had good prognostic power for OS and loco-regional recurrence-free survival (LRFS), with C-index of 0.68 and 0.72, respectively. In all the cases, the addition of radiomics to clinical features improved the prognostic performance. Radiomic features can provide independent prognostic information in NPC patients from non-endemic areas.

The purpose of this study was to establish a methodology and technology for the development of an MRI-based radiomic signature for prognosis of overall survival (OS) in nasopharyngeal cancer from non-endemic areas. The signature was trained using 1072 features extracted from the main tumor in T1-weighted and T2-weighted images of 142 patients. A model with 2 radiomic features was obtained (RAD model). Tumor volume and a signature obtained by training the model on permuted survival data (RADperm model) were used as a reference. A 10-fold cross-validation was used to validate the signature. Harrel's C-index was used as performance metric. A statistical comparison of the RAD, RADperm and volume was performed using Wilcoxon signed rank tests. The C-index for the RAD model was higher compared to the one of the RADperm model (0.69±0.08 vs 0.47±0.05), which ensures absence of overfitting. Also, the signature obtained with the RAD model had an improved C-index compared to tumor volume alone (0.69±0.08 vs 0.65±0.06), suggesting that the radiomic signature provides additional prognostic information.

The three-dimensional (3D) representation of the bifurcation anatomy and disease burden is essential for better understanding of the anatomical complexity of bifurcation disease and planning of stenting strategies. We propose a novel methodology for 3D reconstruction of coronary artery bifurcations based on the integration of angiography, which provides the backbone of the bifurcation, with optical coherence tomography (OCT), which provides the vessel shape. Our methodology introduces several technical novelties to tackle the OCT frame misalignment, correct positioning of the OCT frames at the carina, lumen surface reconstruction, and merging of bifurcation lumens. The accuracy and reproducibility of the methodology were tested in n = 5 patient-specific silicone bifurcations compared to contrast-enhanced micro-computed tomography (µCT), which was used as reference. The feasibility and time-efficiency of the method were explored in n = 7 diseased patient bifurcations of varying anatomical complexity. The OCT-based reconstructed bifurcation models were found to have remarkably high agreement compared to the µCT reference models, yielding r2 values between 0.91 and 0.98 for the normalized lumen areas, and mean differences of 0.005 for lumen shape and 0.004 degrees for bifurcation angles. Likewise, the reproducibility of our methodology was remarkably high. Our methodology successfully reconstructed all the patient bifurcations yielding favorable processing times (average lumen reconstruction time

In this paper, several radiomics-based predictive models of response to induction chemotherapy (IC) in sinonasal cancers (SNCs) are built and tested. Models were built as a combination of radiomic features extracted from three types of MRI images: T1-weighted images, T2-weighted images and apparent diffusion coefficient (ADC) maps. Fifty patients (aged 54 ± 12 years, 41 men) were included in this study. Patients were classified according to their response to IC (25 responders and 25 nonresponders). Not all types of images were acquired for all of the patients: 49 had T1-weighted images, 50 had T2-weighted images and 34 had ADC maps. Only in a subset of 33 patients were all three types of image acquired. Eighty-nine radiomic features were extracted from the MRI images. Dimensionality reduction was performed by using principal component analysis (PCA) and by selecting only the three main components. Different algorithms (trees ensemble, K-nearest neighbors, support vector machine, naive Bayes) were used to classify the patients as either responders or nonresponders. Several radiomic models (either monomodality or multimodality obtained by a combination of T1-weighted, T2-weighted and ADC images) were developed and the performance was assessed through 100 iterations of train and test split. The area under the curve (AUC) of the models ranged from 0.56 to 0.78. Trees ensemble, support vector machine and naive Bayes performed similarly, but in all cases ADC-based models performed better. Trees ensemble gave the highest AUC (0.78 for the T1-weighted+T2-weighted+ADC model) and was used for further analyses. For trees ensemble, the models based on ADC features performed better than those models that did not use those features (P < 0.02 for one-tail Hanley test, AUC range 0.68-0.78 vs 0.56-0.69) except the T1-weighted+ADC model (AUC 0.71 vs 0.69, nonsignificant differences). The results suggest the relevance of ADC-based radiomics for prediction of response to IC in SNCs.

The objectives of the study are to develop a new way to assess stability and discrimination capacity of radiomic features without the need of test-retest or multiple delineations and to use information obtained to perform a preliminary feature selection. Apparent diffusion coefficient (ADC) maps were computed from diffusion-weighted magnetic resonance images (DW-MRI) of two groups of patients: 18 with soft tissue sarcomas (STS) and 18 with oropharyngeal cancers (OPC). Sixty-nine radiomic features were computed, using three different histogram discretizations (16, 32, and 64 bins). Geometrical transformations (translations) of increasing entity were applied to the regions of interest (ROIs), and the intra-class correlation coefficient (ICC) was used to compare the features computed on the original and modified ROIs. The distribution of ICC values for minimal and maximal entity translations (ICC10 and ICC100, respectively) was used to adjust thresholds of ICC (ICCmin and ICCmax) used to discriminate between good, unstable (ICC10 < ICCmin), and non-discriminative features (ICC100 > ICCmax). Fifty-four and 59 radiomic features passed the stability-based selection for all the three histogram discretizations for the OPC and STS datasets, respectively. The excluded features were similar across the different histogram discretizations (Jaccard’s index 0.77 ± 0.13 and 0.9 ± 0.1 for OPC and STS, respectively) but different between datasets (Jaccard’s index 0.19 ± 0.02). The results suggest that the observed radiomic features are mainly stable and discriminative, but the stability depends on the region of the body under observation. The method provides a way to assess stability without the need of test-retest or multiple delineations. Electronic supplementary material The online version of this article (10.1007/s10278-018-0092-9) contains supplementary material, which is available to authorized users.

Centerlines of blood vessels are useful tools to make important anatomical measurements (length, diameter, area), which cannot be accurately obtained using 2D images. In this paper a brand new method for centerline extraction of vascular trees is presented. By using computational fluid dynamics (CFD) we are able to obtain a robust and purely functional centerline allowing us to support better measurements than classic purely geometrical-based centerlines. We show that the CFD-based centerline is within a few pixels from the geometrical centerline where the latter is defined (far away from inlet/outlets and from the branches). We show that the centerline computed with our method is not affected by traditional errors of other classical volume-based algorithms such as topological thinning, and could be a potential alternative to be considered for future studies.

Survival of pediatric patients with brain tumor has increased over the past 20 years, and increasing evidence of iatrogenic toxicities has been reported. In follow-ups, images are acquired at different time points where substantial changes of brain morphology occur, due to childhood physiological development and treatment effects. To address the image registration complexity, we propose two multi-metric approaches (Mplus, Mdot), combining mutual information (MI) and normalized gradient field filter (NGF). The registration performance of the proposed metrics was assessed on a simulated dataset (Brainweb) and compared with those obtained by MI and NGF separately, using mean magnitude and mean angular errors. The most promising metric (Mplus) was then selected and tested on a retrospective dataset comprising 45 pediatric patients who underwent focal radiotherapy for brain cancer. The quality of the realignment was scored by a radiation oncologist using a perceived misalignment metric (PM). All patients but one were assessed as PM ≤ 2 (good alignment), but the remaining one, severely affected by hydrocephalus and pneumocephalus at the first MRI acquisition, scored PM = 5 (unacceptable). These preliminary findings suggest that Mplus might improve the registration accuracy in complex applications such as pediatric oncology, when data are acquired throughout the years of follow-up, and is worth investigating.

Patients with peripheral artery disease who undergo endovascular treatment are often inflicted by in-stent restenosis. The relation between restenosis and abnormal hemodynamics may be analyzed using patient-specific computational fluid dynamics (CFD) simulations. In this work, first a three-dimensional (3D) reconstruction method, based on an in-house semi-automatic segmentation algorithm of a patient’s computed tomography (CT) images with calcification and metallic artifacts, and thrombus removal is de- scribed. The reconstruction method was validated using 3D printed rigid phantoms of stented femoral arteries by comparing the reconstructed geometries with the reference computer-aided design (CAD) ge- ometries employed for 3D printing. The mean reconstruction error resulting from the validation of the re- construction method was ~6% in both stented and non-stented regions. Secondly, a patient-specific model of the stented femoral artery was created and CFD analyses were performed with emphasis on the se- lection of the boundary conditions. CFD results were compared in scenarios with and without common femoral artery bifurcation, employing flat or parabolic inlet velocity profiles. Similar helical flow struc- tures were visible in all scenarios. Negligible differences in wall shear stress (< 0.5%) were found in the stented region. In conclusion, a robust method, applicable to patient-specific cases of stented diseased femoral arteries, was developed and validated.

The rupture of a vulnerable plaque, ulceration, is the most common cause of myocardial infarction. It can be recognized by angiographic features, such as prolonged intraluminal filling and delayed clearance of the contrast liquid. The diagnosis of such an event is an open challenge due to the limited angiographic resolution and acquisition frequency. The treatment of ulcerated plaques is an open discussion, due to the high heterogeneity and the lack of evidences that support particular strategies. Therefore, the therapeutic decision should follow a detailed investigation with angiography and intravascular imaging, such as optical coherence tomography (OCT), to locate the lesion, besides its geometric features and the lumen occlusion severity. The aim of this study is the application of a framework for the in-silico analysis of the disrupted hemodynamics due to an ulcerated lesion. The study employed a validated OCT-based reconstruction methodology and computational fluid dynamics (CFD) simulations for the computation of local hemodynamic quantities, such as wall shear stress. The achieved findings, such as disrupted pre-operative flow conditions, proved the applicability of the developed framework for CFD analyses on complicated patient-specific anatomies that feature ulcerated plaques. The prediction of lesion expansion and the clinical decision making can benefit from a reliable computation of wall shear stress distributions that result from the peculiar anatomy of the lesion. Therefore, the application of intravascular OCT imaging, high fidelity 3D reconstructions and CFD simulations might guide the treatment of such pathology.

Baseline clinical prognostic factors for recurrent and/or metastatic (RM) head and neck squamous cell carcinoma (HNSCC) treated with immunotherapy are lacking. CT-based radiomics may provide additional prognostic information. A total of 85 patients with RM-HNSCC were enrolled for this study. For each tumor, radiomic features were extracted from the segmentation of the largest tumor mass. A pipeline including different feature selection steps was used to train a radiomic signature prognostic for 10-month overall survival (OS). Features were selected based on their stability to geometrical transformation of the segmentation (intraclass correlation coefficient, ICC > 0.75) and their predictive power (area under the curve, AUC > 0.7). The predictive model was developed using the least absolute shrinkage and selection operator (LASSO) in combination with the support vector machine. The model was developed based on the first 68 enrolled patients and tested on the last 17 patients. Classification performance of the radiomic risk was evaluated accuracy and the AUC. The same metrics were computed for some baseline predictors used in clinical practice (volume of largest lesion, total tumor volume, number of tumor lesions, number of affected organs, performance status). The AUC in the test set was 0.67, while accuracy was 0.82. The performance of the radiomic score was higher than the one obtainable with the clinical variables (largest lesion volume: accuracy 0.59, AUC = 0.55; number of tumoral lesions: accuracy 0.71, AUC 0.36; number of affected organs: accuracy 0.47; AUC 0.42; total tumor volume: accuracy 0.59, AUC 0.53; performance status: accuracy 0.41, AUC = 0.47). Radiomics may provide additional baseline prognostic value compared to the variables used in clinical practice.

6077 Background: Antiangiogenic tyrosine kinase inhibitors (TKIs) represent the first-line treatment for radioiodine-refractory differentiated thyroid cancer (RR-DTC). Currently, no predictive factors for the activity of these drugs are available. We investigated whether radiomics may have a predictive role in this setting. Methods: We retrospectively identified patients (pts) affected by metastatic RR-DTC, treated with TKIs between July 2008 and January 2020 at our Institution, with availability of computed tomography (CT) scans at baseline and after at least 2 courses of TKI. Response to TKIs was evaluated according to RECIST v1.1. Pts with complete or partial response at the first radiological evaluation were considered responders (R), pts with stable or progressive disease non-responders (NR). A dedicated radiologist segmented the target lesions as regions of interest (ROIs). Radiomic features related to multiple categories (shape and size, first order statistics, textural features) were extracted from each ROI and computed using the PyRadiomics library v. 3.0. A semi-supervised form of principal component analysis estimated principal components that were then used for response classification through a k nearest neighbors (kNN) classifier. The quality of the model was assessed through train-validation-test split (55% of the data used as training set, 25% as validation set, 20% as test set), repeated 100 times. Performance of the predictive models was quantified with the mean Area Under the ROC Curve (AUC) obtained in the test set. Results: A total of 51 pts with metastatic RR-DTC who had received lenvatinib (n=37), sorafenib (n=4), axitinib (n=3), or vandetanib (n=7) were analyzed. Median age was 64.6 years, with a male prevalence (72.5%). Metastatic sites were lung (84.3%), bone (35.3%), brain (9.9%). Median time from TKI treatment start to the first radiological evaluation was 2.77 months, 24 pts (47%) were R (all partial responses) and 27 (52.9%) NR. In the radiomic analysis, 851 features were computed and 4-19 principal components were selected. Models’ performance of prediction of early response to TKIs is presented in Table. For each value of AUC, the corresponding 95% confidence interval is reported in brackets. Conclusions: Radiomics predicted the response to TKIs of RR-DTC pts with an accuracy of 71%. Radiomics technique has the potential to enable clinicians to anticipate the probability of response to TKIs at baseline, directing toward the most suitable patient-tailored therapeutic path. Prospective studies may further validate these preliminary findings.[Table: see text]

6553 Background: Emerging data suggest that radiomics can be used to predict outcomes in SCCHN. At present, only few data are available for pre-treatment MRI. Methods: Study population was retrieved from an ongoing multicenter, randomized, prospective trial (NCT02262221, HETeCo) evaluating health and economic outcomes of two different follow-up (FUP) strategies (intensive vs non-intensive) in effectively cured stage III-IV (VIII TNM ed.) SCCHN. We selected only patients with both pre- and post-contrast enhancement T1 and T2-weighted baseline MRI (b-MRI) and at least 2 years (2y) of FUP. A radiomic model was developed to identify high risk (HR) and low risk (LR) of disease recurrence. Radiomic features (RF) were extracted from the primary tumor in the b-MRI. The best RF combination was selected by Least Absolute Shrinkage and Selection Operator (LASSO). Ten-fold cross-validation was used to compute sensitivity, specificity and area under the curve (AUC) of the classifier. Kaplan-Meier (KM) curves were estimated for HR and LR, for both overall survival (OS) and disease-free survival (DFS) and log rank test was performed. Three years (3y)-DFS and OS were also estimated for the two groups. The radiomic risk class was used as a new variable in a multivariate Cox model including well established prognostic factors in SCCHN (TNM stage, subsite and HPV). Results: Out of 155 enrolled HETeCO patients, 98 baseline imaging were retrieved of which 57 b-MRI. Of these, 51 met the eligibility criteria (25 in intensive and 26 in non-intensive arm). Baseline patients’ characteristics were: median age 66 yr (38-86); sex (M 42; F 9); median smoking history: 30 packs/y (1-100); 25 oral cavity (49%), 18 oropharynx (35%, 14 HPV+), 6 larynx (12%), 2 hypopharynx (4%). At a median FUP of 42 months (25-64), 45 (88%) patients are still alive. The recurrence rate was 20% (10/51, of which 2 distant). In total, 1608 RF were extracted. The sensitivity, specificity and AUC of the classifier were 90%, 76%, and 80%, respectively. The radiomic risk class was found to be an independent prognostic factor for both DFS and OS (p=0.01 and p=0.046, respectively). KM curves for DFS and OS were significantly different between HR and LR groups (p=0.002 and p=0.04, respectively). In HR vs LR, 3-y DFS and OS were: 78% [61-100%] vs 97% [90-100%], and 88% [75-100%] vs 96% [88-100%], respectively. Conclusions: Radiomics of pre-treatment MRI can predict outcomes in SCCHN. External validation of this preliminary radiomics-based model is currently ongoing.

e23532 Background: Prognosis of extremity soft tissue sarcomas (ESTS) and retroperitoneal sarcomas (RPS) is currently estimated on clinical-pathological features, as those incorporated in the Sarculator nomogram. This study investigated the added value of end-point specific radiomic signatures (RS) in patients with primary ESTS and RPS. Methods: We retrospectively analysed data from patients with a ESTS (N = 100) or a RPS (N = 116) who underwent surgery (Jan2011-Dec2015) at one sarcoma reference institute and had magnetic resonance imaging (MRI) or computed tomography (CT) scan, respectively. Radiomic features were extracted from DICOM files of MRI and CT scan for ESTS and RPS, respectively. The study end-points were incidence of distant metastases (DM) and overall survival (OS) for ESTS and disease-free survival (DFS) and OS for RPS. A RS was established for each end-point. Models discriminative ability was measured using the Harrell c-index (C). Results: ESTS histology was undifferentiated pleomorphic sarcoma, myxofibrosarcoma, myxoid liposarcoma, dedifferentiated/pleomorphic liposarcoma, and others in 28, 21, 18, 9, and 24 patients, respectively. At a median follow-up of 61 months, the 5-yr DM-free rate and OS were 73.6% (95%CI 64.2-82.2%) and 73.4% (95%CI 64.5-83.6), respectively. A prognostic model based on the Sarculator and a DM-RS achieved higher C (0.86, 95%CI 0.77-0.93) than the Sarculator alone (0.83, 95%CI 0.71-0.94). For OS, the RS was not able to increase the Sarculator discriminative ability (Sarculator and OS-RS: C = 0.73, 95%CI 0.62-0.82; Sarculator alone: C = 0.72, 95%CI 0.62-0.81). RPS histology was well differentiated liposarcoma, dedifferentiated liposarcoma, leiomyosarcoma, and other in 22 (19%), 60 (52%), 22 (19%), and 12 (10%) patients, respectively. At a median follow-up of 58 months, 5-yr DFS and OS were 46.7% (95%CI, 37.8-57.6%) and 68% (95% CI 59.9-79.0%), respectively. RS extracted from CT scan without contrast enhancement slightly improve the discriminative ability of the Sarculator for DFS (Sarculator and DFS-RS: C = 0.65, 95%CI 0.57-0-72; Sarculator alone: C = 0.63, 95%CI 0.55-0.70) but not for OS (0.70, 95%CI 0.60-0.80 and 0.69, 95%CI 0.59-0.79, respectively). Conclusions: RS did not improve the discriminative ability of the Sarculator for OS. The added value for RS in relapse prediction (DM incidence for ESTS and DFS for RPS) deserves further investigations.

The purpose of this study is to identify a set of radiomic features extracted from apparent diffusion coefficient (ADC) maps, obtained using baseline diffusion weighted magnetic resonance imaging (DW-MRI), which are able to predict the outcome of induction chemotherapy (IC) in sinonasal cancers. Such prediction could help the clinician defining the better treatment for a particular patient. Eighty-eight radiomic features were extracted from the ADC maps of 15 patients that underwent IC. A preliminary filtering of the features was made by assessing their stability to geometrical transformations of the region of interest (ROI). Mann-Whitney tests corrected for control of false discoveries were performed to identify the features that could discriminate between responsive and nonresponsive patients (4 and 11 respectively). Twenty features were found to be able to discriminate the two groups and they can potentially be used for prediction of response to treatment.

We present an application of a validated reconstruction methodology for the comparison between patient-specific hemodynamics and neo-intimal thickening at nine months from the intervention. (1) Background: Coronary bifurcation stenting alters the vessel geometry, influencing the local hemodynamics. The evaluation of wall shear stress (WSS) relies on the application of computational fluid dynamics to model its distribution along the coronary tree. The endothelium actively responds to WSS, which triggers eventual cell proliferation to cover the stent struts. (2) Methods: Baseline optical coherence tomography and angiographic data were combined to reconstruct a patient-specific coronary bifurcation with an implanted bioresorbable scaffold and to simulate the hemodynamics. Results were linked with the neo-intimal thickening after nine months from the intervention. (3) Results: Blood velocity patterns were disrupted at the bifurcation due to the presence of the stent. It was observed that 55.6% of the scaffolded lumen surface was exposed to values of time-averaged WSS lower than 0.4 Pa. Follow-up images showed a luminal narrowing of 19% in the main branch. There was also a complete coverage in 99% of struts. (4) Conclusions: This approach provided valuable complementary information that might improve the clinical outcomes in this subset of coronary diseases.

Optical coherence tomography (OCT) is an established catheter-based imaging modality for the assessment of coronary artery disease and the guidance of stent placement during percutaneous coronary intervention. Manual analysis of large OCT datasets for vessel contours or stent struts detection is time-consuming and unsuitable for real-time applications. In this study, a fully automatic method was developed for detection of both vessel contours and stent struts. The method was applied to in vitro OCT scans of eight stented silicone bifurcation phantoms for validation purposes. The proposed algorithm comprised four main steps, namely pre-processing, lumen border detection, stent strut detection, and three-dimensional point cloud creation. The algorithm was validated against manual segmentation performed by two independent image readers. Linear regression showed good agreement between automatic and manual segmentations in terms of lumen area (r>0.99). No statistically significant differences in the number of detected struts were found between the segmentations. Mean values of similarity indexes were >95% and >85% for the lumen and stent detection, respectively. Stent point clouds of two selected cases, obtained after OCT image processing, were compared to the centerline points of the corresponding stent reconstructions from micro computed tomography, used as ground-truth. Quantitative comparison between the corresponding stent points resulted in median values of ~150 μm and ~40 μm for the total and radial distances of both cases, respectively. The repeatability of the detection method was investigated by calculating the lumen volume and the mean number of detected struts per frame for seven repeated OCT scans of one selected case. Results showed low deviation of values from the median for both analyzed quantities. In conclusion, this study presents a robust automatic method for detection of lumen contours and stent struts from OCT as supported by focused validation against both manual segmentation and micro computed tomography and by good repeatability.

Radiomics extracts a large number of features from medical images to perform a quantitative characterization. Aim of this study was to assess radiomic features stability and relevance. Apparent diffusion coefficient (ADC) maps were computed from diffusion-weighted magnetic resonance images (DW-MRI) of 18 patients diagnosed with soft-tissue sarcomas (STSs). Thirty-seven intensity-based features were computed on the regions of interest (ROIs). First, ROIs of the images were subjected to translations and rotations in specific ranges. The 37 features computed on the original and transformed ROIs were compared in terms of percentage of variations. The intra-class correlation coefficient (ICC) was computed. To be accepted, a feature should satisfy the following conditions: the ICC after a minimum entity transformation is > 0.6 and the ICC after a maximum entity translation is < 0.4. In total, 31 features out of 37 were accepted by the algorithm. This stability analysis can be used as a first step in the features selection process.

The clinical challenge of percutaneous coronary interventions (PCI) is highly dependent on the recognition of the coronary anatomy of each individual. The classic imaging modality used for PCI is angiography, but advanced imaging techniques that are routinely performed during PCI, like optical coherence tomography (OCT), may provide detailed knowledge of the pre-intervention vessel anatomy as well as the post-procedural assessment of the specific stent-to-vessel interactions. Computational fluid dynamics (CFD) is an emerging investigational tool in the setting of optimization of PCI results. In this study, an OCT-based reconstruction method was developed for the execution of CFD simulations of patient-specific coronary artery models which include the actual geometry of the implanted stent. The method was applied to a rigid phantom resembling a stented segment of the left anterior descending coronary artery. The segmentation algorithm was validated against manual segmentation. A strong correlation was found between automatic and manual segmentation of lumen in terms of area values. Similarity indices resulted >96% for the lumen segmentation and >77% for the stent strut segmentation. The 3D reconstruction achieved for the stented phantom was also assessed with the geometry provided by X-ray computed micro tomography scan, used as ground truth, and showed the incidence of distortion from catheter-based imaging techniques. The 3D reconstruction was successfully used to perform CFD analyses, demonstrating a great potential for patient-specific investigations. In conclusion, OCT may represent a reliable source for patient-specific CFD analyses which may be optimized using dedicated automatic segmentation algorithms.

Purpose The purpose of the paper was to use a virtual phantom to identify a set of radiomic features from T1-weighted and T2-weighted magnetic resonance imaging (MRI) of the brain which is stable to variations in image acquisition parameters and to evaluate the effect of image preprocessing on radiomic features stability. Methods Stability to different sources of variability (time of repetition and echo, voxel size, random noise and intensity non-uniformity) was evaluated for both T1-weighted and T2-weighted MRI images. A set of 107 radiomic features, accounting for shape and size, first order statistics, and textural features was used. Feature stability was quantified using intraclass correlation coefficient (ICC). For each source of variability, stability was evaluated before and after preprocessing (Z-score normalization, resampling, gaussian filtering and bias field correction). Features that have ICC > 0.75 in all the analysis of variability are selected as stable features. Last, the robust feature sets were tested on images acquired with random simulation parameters to assess their generalizability to unseen conditions. Results Preprocessing significantly increased the robustness of radiomic features to the different sources of variability. When preprocessing is applied, a set of 67 and 61 features resulted as stable for T1-weighted and T2-wieghted images respectively, over 80% of which were confirmed by the analysis on the images acquired with random simulation parameters. Conclusion A set of MRI-radiomic features, robust to changes in TR/TE/PS/ST, was identified. This set of features may be used in radiomic analyses based on T1-weighted and T2-weighted MRI images.

Background Despite advances in treatments, 30% to 50% of stage III-IV head and neck squamous cell carcinoma (HNSCC) patients relapse within 2 years after treatment. The Big Data to Decide (BD2Decide) project aimed to build a database for prognostic prediction modeling.Methods Stage III-IV HNSCC patients with locoregionally advanced HNSCC treated with curative intent (1537) were included. Whole transcriptomics and radiomics analyses were performed using pretreatment tumor samples and computed tomography/magnetic resonance imaging scans, respectively.Results The entire cohort was composed of 71% male (1097)and 29% female (440): oral cavity (429, 28%), oropharynx (624, 41%), larynx (314, 20%), and hypopharynx (170, 11%); median follow-up 50.5 months. Transcriptomics and imaging data were available for 1284 (83%) and 1239 (80%) cases, respectively; 1047 (68%) patients shared both.Conclusions This annotated database represents the HNSCC largest available repository and will enable to develop/validate a decision support system integrating multiscale data to explore through classical and machine learning models their prognostic role.