Expert Delineations Research Articles

A deep learning segmentation method to assess dose to organs at risk during breast radiotherapy

Background and purposeContouring of organs at risk is important for studying health effects following breast radiotherapy. However, manual contouring is time-consuming and subject to variability. The purpose of this study was to develop a deep learning-based method to automatically segment multiple structures on breast radiotherapy planning computed tomography (CT) images. Materials and methodsWe used data from 118 patients, including 90 diagnostic CT scans with expert structure delineations for training and 28 breast radiotherapy planning CT images for testing. The radiotherapy CT images also had expert delineations for evaluating performance. We targeted a total of eleven organs at risk including five heart substructures. Segmentation performance was evaluated using the metrics of Dice similarity coefficient (DSC), overlap fraction, volume similarity, Hausdorff distance, mean surface distance, and dose. ResultsThe average DSC achieved on the radiotherapy planning images was 0.94 ± 0.02 for the whole heart, 0.96 ± 0.02 and 0.97 ± 0.01 for the left and right lung, 0.61 ± 0.10 for the esophagus, 0.81 ± 0.04 and 0.86 ± 0.04 for left and right atrium, 0.91 ± 0.02 and 0.84 ± 0.04 for left and right ventricle, and 0.21 ± 0.11 for the left anterior descending artery (LAD), respectively. Except for the LAD, the median difference in mean dose to these structures was small with absolute (relative) differences < 0.1 Gy (6 %). ConclusionsExcept for the LAD, our method demonstrated excellent performance and can be generalized to segment additional structures of interest.

CHELSA-TraCE21k – high-resolution (1 km) downscaled transient temperature and precipitation data since the Last Glacial Maximum

Abstract. High-resolution, downscaled climate model data are used in a wide variety of applications across environmental sciences. Here we introduce a new, high-resolution dataset, CHELSA-TraCE21k. It is obtained by downscaling TraCE-21k data, using the “Climatologies at high resolution for the earth's land surface areas” (CHELSA) V1.2 algorithm with the objective to create global monthly climatologies for temperature and precipitation at 30 arcsec spatial resolution in 100-year time steps for the last 21 000 years. Paleo-orography at high spatial resolution and for each time step is created by combining high-resolution information on glacial cover from current and Last Glacial Maximum (LGM) glacier databases and interpolations using data from a global model of glacial isostasy (ICE-6G_C) and a coupling to mean annual temperatures from TraCE21k (Transient Climate Evolution of the last 21 000 years) based on the Community Climate System Model version 3 (CCSM3). Based on the reconstructed paleo-orography, mean annual temperature and precipitation were downscaled using the CHELSA V1.2 algorithm. The data were validated by comparisons with the glacial extent of the Laurentide ice sheet based on expert delineations, proxy data from Greenland ice cores, historical climate data from meteorological stations, and a dynamic simulation of species distributions throughout the Holocene. Validations show that the CHELSA-TraCE21k V1.0 dataset reasonably represents the distribution of temperature and precipitation through time at an unprecedented 1 km spatial resolution, and simulations based on the data are capable of detecting known LGM refugia of species.

A new artificial intelligence system successfully detectsand localises early neoplasia in Barrett's esophagus by using convolutional neural networks.

Background and aimsSeattle protocol biopsies for Barrett's Esophagus (BE) surveillance are labour intensive with low compliance. Dysplasia detection rates vary, leading to missed lesions. This can potentially be offset with computer aided detection. We have developed convolutional neural networks (CNNs) to identify areas of dysplasia and where to target biopsy.Methods119 Videos were collected in high‐definition white light and optical chromoendoscopy with i‐scan (Pentax Hoya, Japan) imaging in patients with dysplastic and non‐dysplastic BE (NDBE). We trained an indirectly supervised CNN to classify images as dysplastic/non‐dysplastic using whole video annotations to minimise selection bias and maximise accuracy. The CNN was trained using 148,936 video frames (31 dysplastic patients, 31 NDBE, two normal esophagus), validated on 25,161 images from 11 patient videos and tested on 264 iscan‐1 images from 28 dysplastic and 16 NDBE patients which included expert delineations. To localise targeted biopsies/delineations, a second directly supervised CNN was generated based on expert delineations of 94 dysplastic images from 30 patients. This was tested on 86 i‐scan one images from 28 dysplastic patients.FindingsThe indirectly supervised CNN achieved a per image sensitivity in the test set of 91%, specificity 79%, area under receiver operator curve of 93% to detect dysplasia. Per‐lesion sensitivity was 100%. Mean assessment speed was 48 frames per second (fps). 97% of targeted biopsy predictions matched expert and histological assessment at 56 fps. The artificial intelligence system performed better than six endoscopists.InterpretationOur CNNs classify and localise dysplastic Barrett's Esophagus potentially supporting endoscopists during surveillance.

Linked color imaging improves identification of early gastric cancer lesions by expert and non-expert endoscopists

Background and aimsEarly gastric cancer (EGC) lesions are often subtle and endoscopically poorly visible. The aim of this study is to evaluate the additive effect of linked color imaging (LCI) next to white-light endoscopy (WLE) for identification of EGC, when assessed by expert and non-expert endoscopists.MethodsForty EGC cases were visualized in corresponding WLE and LCI images. Endoscopists evaluated the cases in 3 assessment phases: Phase 1: WLE images only; Phase 2: LCI images only; Phase 3: WLE and LCI images side-to-side. First, 3 expert endoscopists delineated all cases. A high level of agreement between the expert delineations corresponded with a high AND/OR ratio. Subsequently, 62 non-experts indicated their preferred biopsy location. Outcomes of the study are as follows: (1) difference in expert AND/OR ratio; (2) accuracy of biopsy placement by non-expert endoscopists; and (3) preference of imaging modality by non-expert endoscopists.ResultsQuantitative agreement between experts increased significantly when LCI was available (0.58 vs. 0.46, p = 0.007). This increase was more apparent for the more challenging cases (0.21 vs. 0.47, p < 0.001). Non-experts placed the biopsy mark more accurately with LCI (82.3% vs. 87.2%, p < 0.001). Again this increase was more profound for the more challenging cases (70.4% vs. 83.4%, p < 0.001). Non-experts indicated to prefer LCI over WLE.ConclusionThe addition of LCI next to WLE improves visualization of EGC. Experts reach higher consensus on discrimination between neoplasia and inflammation when using LCI. Non-experts improve their targeted biopsy placement with the use of LCI. LCI therefore appears to be a useful tool for identification of EGC.Graphical abstract

Simple delineations cannot substitute full 3d tumor delineations for MR-based radiomics prediction of locoregional control in oropharyngeal cancer

Background and purposeManual delineation of head and neck tumor contours for radiomics analyses is tedious and time consuming. This study investigates if fast or readily available tumor contours can substitute full tumor contours by an experienced observer for an MR-based radiomics model to predict locoregional control (LRC) in oropharyngeal squamous cell carcinoma (OPSCC) tumors. Materials and methodsRadiomic features were extracted from postcontrast T1-weighted MRIs of 177 OPSCC primary tumors using six different manual delineation strategies. LRC prediction models based on recursive feature elimination combined with logistic regression were built. Models were trained and tested on data from each separate delineation. Additionally, the model derived from segmentations from the experienced reader was tested by each of the alternative delineations. Complementary, this was repeated with removal of size and shape features. Model performance was evaluated using area under the curve (AUC). ResultsPrediction performance of the experienced radiologist tumor delineation (AUC: 0.74) was superior compared to all other delineations when trained and tested (AUCs: 0.41–0.56) or trained on experienced delineations and tested (AUC: 0.56–0.67) on alternative segmentations. Removal of size and shape features considerably decreases prediction performance (AUC: 0.54). Applying the model based on expert delineations to spherical or single slice delineations makes prediction worthless since these models predict one class. ConclusionFast or readily available contours cannot substitute full expert tumor delineations in radiomics models predictive of LRC in OPSCC.

Myocardial Infarction Quantification from Late Gadolinium Enhancement MRI Using Top-Hat Transforms and Neural Networks

Late gadolinium enhancement (LGE) MRI is the gold standard technique for myocardial viability assessment. Although the technique accurately reflects the damaged tissue, there is no clinical standard to quantify myocardial infarction (MI). Moreover, commercial software used in clinical practice are mostly semi-automatic, and hence require direct intervention of experts. In this work, a new automatic method for MI quantification from LGE-MRI is proposed. Our novel segmentation approach is devised for accurately detecting not only hyper-enhanced lesions, but also microvascular obstruction areas. Moreover, it includes a myocardial disease detection step which extends the algorithm for working under healthy scans. The method is based on a cascade approach where firstly, diseased slices are identified by a convolutional neural network (CNN). Secondly, by means of morphological operations a fast coarse scar segmentation is obtained. Thirdly, the segmentation is refined by a boundary-voxel reclassification strategy using an ensemble of very light CNNs. We tested the method on a LGE-MRI database with healthy (n = 20) and diseased (n = 80) cases following a 5-fold cross-validation scheme. Our approach segmented myocardial scars with an average Dice coefficient of 77.22 ± 14.3% and with a volumetric error of 1.0 ± 6.9 cm3. In a comparison against nine reference algorithms, the proposed method achieved the highest agreement in volumetric scar quantification with the expert delineations (p&lt; 0.001 when compared to the other approaches). Moreover, it was able to reproduce the scar segmentation intra- and inter-rater variability. Our approach was shown to be a good first attempt towards automatic and accurate myocardial scar segmentation, although validation over larger LGE-MRI databases is needed.

Validation of separate multi-atlases for auto segmentation of cardiac substructures in CT-scans acquired in deep inspiration breath hold and free breathing

Background and purposeDeveloping NTCP-models for cardiac complications after breast cancer (BC) radiotherapy requires cardiac dose-volume parameters for many patients. These can be obtained by using multi-atlas based automatic segmentation (MABAS) of cardiac structures in planning CT scans. We investigated the relevance of separate multi-atlases for deep inspiration breath hold (DIBH) and free breathing (FB) CT scans. Materials and methodsBC patients scanned in DIBH (n = 10) and in FB (n = 20) were selected to create separate multi-atlases consisting of expert panel delineations of the whole heart, atria and ventricles. The accuracy of atlas-generated contours was validated with expert delineations in independent datasets (n = 10 for DIBH and FB) and reported as Dice coefficients, contour distances and dose-volume differences in relation to interobserver variability of manual contours. Dependency of MABAS contouring accuracy on breathing technique was assessed by validation of a FB atlas in DIBH patients and vice versa (cross-validation). ResultsFor all structures the FB and DIBH atlases resulted in Dice coefficients with their respective reference contours ≥ 0.8 and average contour distances ≤ 2 mm smaller than slice thickness of (CTs). No significant differences were found for dose-volume parameters in volumes receiving relevant dose levels (WH, LV and RV). Accuracy of the DIBH atlas was at least similar to, and for the ventricles better than, the interobserver variation in manual delineation. Cross-validation between breathing techniques showed a reduced MABAS performance. ConclusionMulti-atlas accuracy was at least similar to interobserver delineation variation. Separate atlases for scans made in DIBH and FB could benefit atlas performance because accuracy depends on breathing technique.

Quality assurance of radiotherapy in the ongoing EORTC 1420 \u201cBest of\u201d trial for early stage oropharyngeal, supraglottic and hypopharyngeal carcinoma: results of the benchmark case procedure

IntroductionThe current phase III EORTC 1420 Best-of trial (NCT02984410) compares the swallowing function after transoral surgery versus intensity modulated radiotherapy (RT) in patients with early-stage carcinoma of the oropharynx, supraglottis and hypopharynx. We report the analysis of the Benchmark Case (BC) procedures before patient recruitment with special attention to dysphagia/aspiration related structures (DARS).Materials and methodsSubmitted RT volumes and plans from participating centers were analyzed and compared against the gold-standard expert delineations and dose distributions. Descriptive analysis of protocol deviations was conducted. Mean Sorensen-Dice similarity index (mDSI) and Hausdorff distance (mHD) were applied to evaluate the inter-observer variability (IOV).Results65% (23/35) of the institutions needed more than one submission to achieve Quality assurance (RTQA) clearance. OAR volume delineations were the cause for rejection in 53% (40/76) of cases. IOV could be improved in 5 out of 12 OARs by more than 10 mm after resubmission (mHD). Despite this, final IOV for critical OARs in delineation remained significant among DARS by choosing an aleatory threshold of 0.7 (mDSI) and 15 mm (mHD).ConclusionsThis is to our knowledge the largest BC analysis among Head and neck RTQA programs performed in the framework of a prospective trial. Benchmarking identified non-common OARs and target delineations errors as the main source of deviations and IOV could be reduced in a significant number of cases after this process. Due to the substantial resources involved with benchmarking, future benchmark analyses should assess fully the impact on patients’ clinical outcome.

Wound 3D Geometrical Feature Estimation Using Poisson Reconstruction

With the aging population of the developed countries, the diseases of affluence become common. One of the consequences of their complications are chronic wounds that affect more and more people. Because the healing of chronic wounds is a lengthy process, its monitoring should be as simple as possible, yet providing accurate and reliable documentation. In this article, we present an image acquisition system and wound surface reconstruction method using several imaging modalities: color photography, thermal imaging, and depth perception. The proposed method is dedicated mainly to wounds located on the limbs where body curvature is large, and its influence on 2D results cannot be neglected. Our approach minimizes the extra effort taken by the medical staff to prepare the wound outline as it is still performed using 2D color photo and then mapped into the 3D space. The method was validated on 29 data sets containing two 3D point clouds from two depth imaging devices (depth camera and stereo camera) as well as 2D color photos and thermal maps. The approach was compared with expert delineations as well as other contemporary methods for wound surface reconstruction presented in the literature. Performed experiments and the obtained results show that the proposed method is statistically concordant with expert delineations performed in 3D.

Improving Radiotherapy Workflow Through Implementation of Delineation Guidelines & AI-Based Annotation

Incidence of breast cancer is increasing as the number of radiotherapy treatments. Delineation of target volumes and organs at risk (OARs) is time-consuming and suffers from heterogeneity. New guidelines have emerged, towards standardizing the process and improving treatment outcomes. The objective of this effort is twofold: assess the benefit of these guidelines and the added value of AI-driven delineation methods in terms of quality and resources gain. A CE-marked solution for automatic delineation of 80+ organs at risk harnessing a unique combination of anatomically preserving and deep learning delineation concept was developed. Using transfer learning the models were re-trained according to the latest ESTRO guidelines, through the integration of 256 cases randomly selected from HYPOG-01 trial. One hundred unseen cases were selected for evaluation: half were delineated based on the ESTRO guidelines (C1) and 50 cases were delineated before guidelines implementation (C2). For each case, automatic delineations (AD) were generated and blended with the ones corresponding to the experts for qualitative and independent evaluation. Overall, 33% of AD structures, 33% manual structures from C1 and 33% manual structures from C2 were scored by 4 radiation oncologist breast experts as A for “No correction required”, B for “Minor correction required” and C for “Major corrections required”. Correction effort towards moving AD to clinically-acceptable target volumes and OARs (heart, lungs, spinal cord, esophagus and thyroid) were measured in C2 by one expert. Assessing benefit of guidelines, significant gain was observed on expert delineations between C1 & C2. Some OARs were not delineated before guidelines implementation such as thyroid in 95% of C2 cases. The delineations of experts were assessed clinically acceptable (A+B) for 93% of C1 cases, while the percentage ramped down to 75% in C2. In terms of expert versus AI, 93% of the automated delineations in C1 & C2 were considered as clinically acceptable (A: 49%; B: 44%), reaching human expertise. All target volumes were better scored with AD (92% and 94% of A+B for breast and nodes for AD vs. 86% and 87% respectively for manual delineations). Spinal cord and lungs were better scored using AD (94% and 96% respectively of A) than manual delineation (76% and 91%). On the contrary, 35% of the AD brachial plexus required major corrections (13% for the manual ones). The mean time to correct an AD case was 2.6±1.9 min (4.3±1.6 min for cases with nodal treatment; 1.8±1.0 min without). This systematic, blinded, random evaluation suggests that using AD in breast cancer has high potential for delineation guidelines propagation, homogenization of practices and time saving. Only minor corrections were required, showing the clinical relevance of the developed software. Evaluation of dosimetric impact of AD is on-going on C2 cohort to validate its major interest in clinical practice.

Deep learning-based auto-segmentation of targets and organs-at-risk for magnetic resonance imaging only planning of prostate radiotherapy.

Magnetic resonance (MR) only radiation therapy for prostate treatment provides superior contrast for defining targets and organs-at-risk (OARs). This study aims to develop a deep learning model to leverage this advantage to automate the contouring process. Six structures (bladder, rectum, urethra, penile bulb, rectal spacer, prostate and seminal vesicles) were contoured and reviewed by a radiation oncologist on axial T2-weighted MR image sets from 50 patients, which constituted expert delineations. The data was split into a 40/10 training and validation set to train a two-dimensional fully convolutional neural network, DeepLabV3+, using transfer learning. The T2-weighted image sets were pre-processed to 2D false color images to leverage pre-trained (from natural images) convolutional layers' weights. Independent testing was performed on an additional 50 patient's MR scans. Performance comparison was done against a U-Net deep learning method. Algorithms were evaluated using volumetric Dice similarity coefficient (VDSC) and surface Dice similarity coefficient (SDSC). When comparing VDSC, DeepLabV3+ significantly outperformed U-Net for all structures except urethra (P < 0.001). Average VDSC was 0.93 ± 0.04 (bladder), 0.83 ± 0.06 (prostate and seminal vesicles [CTV]), 0.74 ± 0.13 (penile bulb), 0.82 ± 0.05 (rectum), 0.69 ± 0.10 (urethra), and 0.81 ± 0.1 (rectal spacer). Average SDSC was 0.92 ± 0.1 (bladder), 0.85 ± 0.11 (prostate and seminal vesicles [CTV]), 0.80 ± 0.22 (penile bulb), 0.87 ± 0.07 (rectum), 0.85 ± 0.25 (urethra), and 0.83 ± 0.26 (rectal spacer). A deep learning-based model produced contours that show promise to streamline an MR-only planning workflow in treating prostate cancer.

Performance evaluation of breast lesion detection systems with expert delineations: a comparative investigation on mammographic images

Performance of computerized diagnostic systems yearning to be approved by medical regulatory bodies must meet the expectations of human experts. Highly accurate lesion segmentation techniques have thus turned out to be an essential part for clinical acceptability of mammography based computer-aided diagnosis systems. The objective of this study is to evaluate the performance of six popular breast tumor detection techniques with manual delineations provided by two experienced radiologists on the mammographic images. In our study, 20 mammographic images from the mini-MIAS database are utilized. For the analysis, input mammographic images are first manually cropped to generate the region of interest (ROI). The ROI images are then pre-processed and segmentation is performed using different techniques, namely: expected maximization, K-means, Fuzzy c-Means (FCM), multilevel thresholding, region growing, and particle swarm optimization. The results were compared against the manual tracings. Among the other five segmentation techniques, FCM achieves the highest Jaccard Index (0.73 ± 0.06) and Dice Similarity Coefficient (0.82 ± 0.08) values. Statistical analysis (t-test, Mann Whitney U test, Wilcoxon test, Chi-Square test, and Kolmogorov–Smirnov test) and graphical analysis (Bland Altman and Regression plots) further prove the stability and reliability of the segmentation methods. Segmentation using FCM demonstrates the most accurate results and can be employed for the detection of breast cancer in the mammographic images. Further, it is concluded that computer-aided lesion detection systems can be used to assist Radiologists in routine clinical practice for the detection of breast tumors in mammographic images.

A modified level set algorithm based on point distance shape constraint for lesion and organ segmentation

PurposeThe segmentation of organs and lesions from medical images is a challenging task due to the presents of noise, intensity inhomogeneity, blurry/weak boundaries. In this paper, a point distance shape constraint is proposed and incorporated in the level set framework for the segmentation of objects with various shapes. MethodsThe proposed shape constraint is a linear combination of the Euclidean distance of user selected points. By selecting different numbers of points, it can generate different shape constraints and therefore is more flexible in dealing with different shapes. Then this shape constraint is incorporated into the variational level set framework. A convex relaxation is applied to get a convex model which can be efficiently solved by a primal-dual hybrid gradient algorithm. ResultsThe proposed algorithm is tested on 60 CT images with the nodular type of hepatic cellular cancer (HCC), 100 ultrasound kidney images, 20 prostate MR images, 20 lumbar CT images and 100 transrectal ultrasound prostate images. The algorithms performance is evaluated using a number of metrics by comparison with expert delineations. The validation results show that, for five datasets mentioned previously, the average DSCs of the proposed algorithm are 95.6% ± 1.4%, 94.3% ± 3.1%, 91.3% ± 3.8%, 92.7% ± 1.5% and 94.4% ± 2.2% respectively. Both quantitative and qualitative evaluation confirm that the proposed method can provide more accurate segmentation than four state-of-the-art methods. ConclusionThe proposed point distance shape constraint segmentation model can accurately segment organs and lesions with a number of shapes in medical images.

Performance evaluation of automated brain tumor detection systems with expert delineations and interobserver variability analysis in diseased patients on magnetic resonance imaging

ABSTRACTIntervention by human expert has turned out to be essential for computerized analysis systems desiring to be approved by medical regulatory bodies. Further, to validate the performance of automated diagnosis systems, interobserver variability analysis is critically important. The purpose of this article is twofold: (i) firstly to perform interobserver variability analysis of two experienced Radiologists interpreting lesion boundary in brain magnetic resonance images; (ii) secondly, to evaluate the performance of recently proposed automated lesion segmentation model with that of the two experienced Radiologists who performed manual delineations of lesion boundary. Experiments were conducted on the database consisting of 80 real-time brain images with glioma tumor acquired using magnetic resonance imaging (MRI). Extensive statistical analysis such as the two tailed T-test, analysis of variance (ANOVA) test, Mann-Whitney U test, regression and correlation tests, etc. are performed to compare the lesions detected manually by experts and that by the automated method. Furthermore, three quantitative measures namely, dice similarity index, Jaccard coefficient, and Hausdorff distance are used to evaluate the automated lesion detection method. The experimental results show that the lesion boundaries detected by the automated method are very close to the manual delineations provided by the expert Radiologists. It is concluded that the automated systems for brain lesion detection can be utilized as a part of routine clinical practice to help the medical professionals in determining the exact location and area of lesions in brain MRI images.

OA049] Tumour segmentation in longitudinal diffusion weighted MRI of rectal cancer patients receiving radiotherapy

Purpose Target volume delineation is a fundamental step in radiotherapy and is subject to considerable inter-observer variation. Diffusion weighted magnetic resonance imaging (DW-MRI) is being investigated for tumour definition in locally advanced rectal cancer (LARC), offering high contrast between volumes with different diffusion characteristics. Segmentation algorithms such as k-means clustering have the potential to improve consistency of target delineation. In this study we have evaluated the performance of two segmentation algorithms for target definition by comparing them to expert delineations on longitudinal DW-MRIs for patients with LARC. Methods Weekly repeat DW-MRI scans (b-values: 0, 150, 1000 s/mm 2 were recorded on a 1.5T Philips Ingenia MR scanner (Best, NL) for five LARC patients treated with neo-adjuvant radiotherapy (50.4 Gy, 28 fractions). Initially, feature scaling was performed on the b = 1000 s/mm 2 image set as well as the calculated Apparent Diffusion Coefficient (ADC) maps. Tumour segmentation was performed using i) k-means clustering and ii) Gaussian mixture segmentation, with three classes reflecting tumour, air and normal tissue. Tumour volumes were selected as the classes having highest mean signal intensity at b = 1000 s/mm 2 and lowest mean ADC value. Manual delineations were performed by two observers until consensus was reached, using primarily the b = 1000 s/mm 2 image set (with the ADC maps taken into consideration). Manual delineations were compared to automatic delineations using the DICE similarity coefficient (DSC), Hausdorff Distance (HD) and mean distance to agreement (DTA). Results For k-means clustering DSC was 0.8 (median; range: 0.1–0.9), HD was 10 mm (median; range: 4.4–26 mm) and DTA was 1.5 mm (median; range: 0.1–0.9 mm). Corresponding results for Gaussian mixture segmentation were 0.7 (median; range: 0.2–0.9), 8.5 mm (median; range: 4.1–28 mm) and 1.7 mm (median; range: 0.8–11 mm). DSC decreased during treatment course for both segmentation algorithms (p Conclusions Tumour segmentation with k-means clustering and Gaussian mixture model based on DW-MRI and ADC maps perform similar and may improve consistency in target volume delineation for LARC patients.

Multiple Resolution Residually Connected Feature Streams for Automatic Lung Tumor Segmentation From CT Images.

Volumetric lung tumor segmentation and accurate longitudinal tracking of tumor volume changes from computed tomography images are essential for monitoring tumor response to therapy. Hence, we developed two multiple resolution residually connected network (MRRN) formulations called incremental-MRRN and dense-MRRN. Our networks simultaneously combine features across multiple image resolution and feature levels through residual connections to detect and segment the lung tumors. We evaluated our method on a total of 1210 non-small cell (NSCLC) lung tumors and nodules from three data sets consisting of 377 tumors from the open-source Cancer Imaging Archive (TCIA), 304 advanced stage NSCLC treated with anti- PD-1 checkpoint immunotherapy from internal institution MSKCC data set, and 529 lung nodules from the Lung Image Database Consortium (LIDC). The algorithm was trained using 377 tumors from the TCIA data set and validated on the MSKCC and tested on LIDC data sets. The segmentation accuracy compared to expert delineations was evaluated by computing the dice similarity coefficient, Hausdorff distances, sensitivity, and precision metrics. Our best performing incremental-MRRN method produced the highest DSC of 0.74 ± 0.13 for TCIA, 0.75±0.12 for MSKCC, and 0.68±0.23 for the LIDC data sets. There was no significant difference in the estimations of volumetric tumor changes computed using the incremental-MRRN method compared with the expert segmentation. In summary, we have developed a multi-scale CNN approach for volumetrically segmenting lung tumors which enables accurate, automated identification of and serial measurement of tumor volumes in the lung.

A Deep Learning Approach for Targeted Contrast-Enhanced Ultrasound Based Prostate Cancer Detection.

The important role of angiogenesis in cancer development has driven many researchers to investigate the prospects of noninvasive cancer diagnosis based on the technology of contrast-enhanced ultrasound (CEUS) imaging. This paper presents a deep learning framework to detect prostate cancer in the sequential CEUS images. The proposed method uniformly extracts features from both the spatial and the temporal dimensions by performing three-dimensional convolution operations, which captures the dynamic information of the perfusion process encoded in multiple adjacent frames for prostate cancer detection. The deep learning models were trained and validated against expert delineations over the CEUS images recorded using two types of contrast agents, i.e., the anti-PSMA based agent targeted to prostate cancer cells and the non-targeted blank agent. Experiments showed that the deep learning method achieved over 91 percent specificity and 90 percent average accuracy over the targeted CEUS images for prostate cancer detection, which was superior ( ) than previously reported approaches and implementations.

MR and CT data with multiobserver delineations of organs in the pelvic area-Part of the Gold Atlas project.

We describe a public dataset with MR and CT images of patients performed in the same position with both multiobserver and expert consensus delineations of relevant organs in the male pelvic region. The purpose was to provide means for training and validation of segmentation algorithms and methods to convert MR to CT like data, i.e., so called synthetic CT (sCT). T1- and T2-weighted MR images as well as CT data were collected for 19 patients at three different departments. Five experts delineated nine organs for each patient based on the T2-weighted MR images. An automatic method was used to fuse the delineations. Starting from each fused delineation, a consensus delineation was agreed upon by the five experts for each organ and patient. Segmentation overlap between user delineations with respect to the consensus delineations was measured to describe the spread of the collected data. Finally, an open-source software was used to create deformation vector fields describing the relation between MR and CT images to further increase the usability of the dataset. The dataset has been made publically available to be used for academic purposes, and can be accessed from https://zenodo.org/record/583096. The dataset provides a useful source for training and validation of segmentation algorithms as well as methods to convert MR to CT-like data (sCT). To give some examples: The T2-weighted MR images with their consensus delineations can directly be used as a template in an existing atlas-based segmentation engine; the expert delineations are useful to validate the performance of a segmentation algorithm as they provide a way to measure variability among users which can be compared with the result of an automatic segmentation; and the pairwise deformably registered MR and CT images can be a source for an atlas-based sCT algorithm or for validation of sCT algorithm.

Coupling Uncertainties with Accuracy Assessment in Object-Based Slum Detections, Case Study: Jakarta, Indonesia

Object-Based Image Analysis (OBIA) has been successfully used to map slums. In general, the occurrence of uncertainties in producing geographic data is inevitable. However, most studies concentrated solely on assessing the classification accuracy and neglecting the inherent uncertainties. Our research analyses the impact of uncertainties in measuring the accuracy of OBIA-based slum detection. We selected Jakarta as our case study area because of a national policy of slum eradication, which is causing rapid changes in slum areas. Our research comprises of four parts: slum conceptualization, ruleset development, implementation, and accuracy and uncertainty measurements. Existential and extensional uncertainty arise when producing reference data. The comparison of a manual expert delineations of slums with OBIA slum classification results into four combinations: True Positive, False Positive, True Negative and False Negative. However, the higher the True Positive (which lead to a better accuracy), the lower the certainty of the results. This demonstrates the impact of extensional uncertainties. Our study also demonstrates the role of non-observable indicators (i.e., land tenure), to assist slum detection, particularly in areas where uncertainties exist. In conclusion, uncertainties are increasing when aiming to achieve a higher classification accuracy by matching manual delineation and OBIA classification.

A multiresolution prostate representation for automatic segmentation in magnetic resonance images.

Accurate prostate delineation is necessary in radiotherapy processes for concentrating the dose onto the prostate and reducing side effects in neighboring organs. Currently, manual delineation is performed over magnetic resonance imaging (MRI) taking advantage of its high soft tissue contrast property. Nevertheless, as human intervention is a consuming task with high intra- and interobserver variability rates, (semi)-automatic organ delineation tools have emerged to cope with these challenges, reducing the time spent for these tasks. This work presents a multiresolution representation that defines a novel metric and allows to segment a new prostate by combining a set of most similar prostates in a dataset. The proposed method starts by selecting the set of most similar prostates with respect to a new one using the proposed multiresolution representation. This representation characterizes the prostate through a set of salient points, extracted from a region of interest (ROI) that encloses theorgan and refined using structural information, allowing to capture main relevant features of the organ boundary. Afterward, the new prostate is automatically segmented by combining the nonrigidly registered expert delineations associated to the previous selected similar prostates using a weighted patch-based strategy. Finally, the prostate contour is smoothed based on morphological operations. The proposed approach was evaluated with respect to the expert manual segmentation under a leave-one-out scheme using two public datasets, obtaining averaged Dice coefficients of 82%±0.07 and 83%±0.06, and demonstrating a competitive performance with respect to atlas-based state-of-the-art methods. The proposed multiresolution representation provides a feature space that follows a local salient point criteria and a global rule of the spatial configuration among these points to find out the most similar prostates. This strategy suggests an easy adaptation in the clinical routine, as supporting tool for annotation.