Tumor Contour Research Articles

SU-C-9A-03: Simultaneous Deconvolution and Segmentation for PET Tumor Delineation Using a Variational Method

Purpose: To implement a new method that integrates deconvolution with segmentation under the variational framework for PET tumor delineation. Methods: Deconvolution and segmentation are both challenging problems in image processing. The partial volume effect (PVE) makes tumor boundaries in PET image blurred which affects the accuracy of tumor segmentation. Deconvolution aims to obtain a PVE-free image, which can help to improve the segmentation accuracy. Conversely, a correct localization of the object boundaries is helpful to estimate the blur kernel, and thus assist in the deconvolution. In this study, we proposed to solve the two problems simultaneously using a variaional method so that they can benefit each other. The energy functional consists of a fidelity term and a regularization term, and the blur kernel was limited to be the isotropic Gaussian kernel. We minimized the energy functional by solving the associated Euler-Lagrange equations and taking the derivative with respect to the parameters of the kernel function. An alternate minimization method was used to iterate between segmentation, deconvolution and blur-kernel recovery. The performance of the proposed method was tested on clinic PET images of patients with non-Hodgkin's lymphoma, and compared with seven other segmentation methods using the dice similarity index (DSI) and volume error (VE). Results: Among all segmentation methods, the proposed one (DSI=0.81, VE=0.05) has the highest accuracy, followed by the active contours without edges (DSI=0.81, VE=0.25), while other methods including the Graph Cut and the Mumford-Shah (MS) method have lower accuracy. A visual inspection shows that the proposed method localizes the real tumor contour very well. Conclusion: The result showed that deconvolution and segmentation can contribute to each other. The proposed variational method solve the two problems simultaneously, and leads to a high performance for tumor segmentation in PET. This work was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 60971112 and 61375018, and Fundamental Research Funds for the Central Universities, under Grant No. 2012QN086. Wei Lu was supported in part by the National Institutes of Health (NIH) Grant No. R01 CA172638.

SU-E-J-224: Multimodality Segmentation of Head and Neck Tumors

Purpose: Develop an algorithm that is able to automatically segment tumor volume in Head and Neck cancer by integrating information from CT, PET and MR imaging simultaneously. Methods: Twenty three patients that were recruited under an adaptive radiotherapy protocol had MR, CT and PET/CT scans within 2 months prior to start of radiotherapy. The patients had unresectable disease and were treated either with chemoradiotherapy or radiation therapy alone. Using the Velocity software, the PET/CT and MR (T1 weighted+contrast) scans were registered to the planning CT using deformable and rigid registration respectively. The PET and MR images were then resampled according to the registration to match the planning CT. The resampled images, together with the planning CT, were fed into a multi-channel segmentation algorithm, which is based on Gaussian mixture models and solved with the expectation-maximization algorithm and Markov random fields. A rectangular region of interest (ROI) was manually placed to identify the tumor area and facilitate the segmentation process. The auto-segmented tumor contours were compared with the gross tumor volume (GTV) manually defined by the physician. The volume difference and Dice similarity coefficient (DSC) between the manual and autosegmented GTV contours were calculated as the quantitative evaluation metrics. Results: The multimodality segmentation algorithm was applied to all 23 patients. The volumes of the auto-segmented GTV ranged from 18.4cc to 32.8cc. The average (range) volume difference between the manual and auto-segmented GTV was −42% (−32.8%–63.8%). The average DSC value was 0.62, ranging from 0.39 to 0.78. Conclusion: An algorithm for the automated definition of tumor volume using multiple imaging modalities simultaneously was successfully developed and implemented for Head and Neck cancer. This development along with more accurate registration algorithms can aid physicians in the efforts to interpret the multitude of imaging information available in radiotherapy today. This project was supported by a grant by Varian Medical Systems.

SU‐E‐J‐207: Assessing the Validity of 4D‐CT Based Target Volumes and Free Breathing CBCT Localization in Lung Stereotactic Ablative Radiation Therapy (SABR)

Purpose:Four‐dimensional‐computed‐tomography(4D‐CT) imaging for target‐volume delineation and cone‐beam‐tomography(CBCT) for treatment localization are widely utilized in lung‐SABR.Aim of this study was to perform a quantitative‐assessment and inter‐comparison of Internal‐targetvolumes(ITV) drawn on various phases of breathing‐cycle 4D‐CT‐scans, Maximum‐intensity‐projection(MIP), average‐intensity‐projection(AIP)and static CT‐scans of lung‐motion‐phantom to simulate lung‐SABR patient geometry. We also analyzed and compared the ITVs drawn on freebreathing‐ CBCT.Materials & Methods:4D‐CT‐scans were acquired on Philips big‐bore 16slice CT and Bellows‐respiratory monitoring‐system using retrospective phase‐binning method. Each respiratory cycle divided into 10‐phases. Quasar‐Phantom with lung‐inserts and 3cm‐diameter nylonball to simulate tumor and was placed on respiratory‐motion‐platform for 4D‐CT and CBCT‐acquisition. Amplitudes of motions: 0.5,1.0,2.0,3.0cm in superior‐inferior direction with breathing‐cycle time of 6,5,4,6sec, respectively used.4D‐CTs with 10‐phases(0%to90%)for each excursion‐set and 3D‐CT for static‐phantom exported to iPlan treatment‐planningsystem(TPS). Tumor‐volumes delineated in all phases of 4D‐CT, MIP,AIP,CBCT scans using fixed‐HU‐threshold(−500to1000)values automatically.For each 4D‐dataset ITV obtained by unifying the tumorcontours on all phases.CBCT‐ITV‐volumes were drawn in Eclipse‐TPS.Results:Mean volume of tumor contours for all phases compared with static 3D‐CT were 0.62±0.08%, 1.67±0.26%, 4.77±0.54% and 9.27±1.23% for 0.5cm,1cm,2cm,3cm excursions respectively. Differences of mean Union‐ITV with MIP‐ITV were close(≤2.4%).Mean Union‐ITV from expected‐theoretical values differed from −4.9% to 3.8%.Union‐ITV and MIP‐ITV were closer within 2.3%. AIP‐ITVs were underestimated from 14 to 32% compared to union‐ITV for all motion datasets. Differences of −5.9% to −44% and −5% to 6.7% for CBCT‐ITV from MIP‐ITV and AIP‐ITV respectively.Motion excursions and centroid positions were within 2mm for 4D‐CT and CBCT‐ITVs to that of expected values.Conclusion:4D‐CT MIP‐ITV and Union‐ITV showed very good agreement that validates that ITV can be fast contoured on MIP.Contouring ITV in AIP must be avoided as it significantly underestimates the volume with all excursions. Free breathing CBCT‐ITV showed good agreement with AIP‐ITV but underestimated the MIP‐ITV. Estimation of excursions and centroid values for 4D‐CT and CBCT were in good agreement with expected values.

Semi-automatic breast ultrasound image segmentation based on mean shift and graph cuts.

Computerized tumor segmentation on breast ultrasound (BUS) images remains a challenging task. In this paper, we proposed a new method for semi-automatic tumor segmentation on BUS images using Gaussian filtering, histogram equalization, mean shift, and graph cuts. The only interaction required was to select two diagonal points to determine a region of interest (ROI) on an input image. The ROI image was shrunken by a factor of 2 using bicubic interpolation to reduce computation time. The shrunken image was smoothed by a Gaussian filter and then contrast-enhanced by histogram equalization. Next, the enhanced image was filtered by pyramid mean shift to improve homogeneity. The object and background seeds for graph cuts were automatically generated on the filtered image. Using these seeds, the filtered image was then segmented by graph cuts into a binary image containing the object and background. Finally, the binary image was expanded by a factor of 2 using bicubic interpolation, and the expanded image was processed by morphological opening and closing to refine the tumor contour. The method was implemented with OpenCV 2.4.3 and Visual Studio 2010 and tested for 38 BUS images with benign tumors and 31 BUS images with malignant tumors from different ultrasound scanners. Experimental results showed that our method had a true positive rate (TP) of 91.7%, a false positive (FP) rate of 11.9%, and a similarity (SI) rate of 85.6%. The mean run time on Intel Core 2.66 GHz CPU and 4 GB RAM was 0.49 ± 0.36 s. The experimental results indicate that the proposed method may be useful in BUS image segmentation.

Prediction of Lung Tumor Evolution During Radiotherapy in Individual Patients With PET

We propose a patient-specific model based on partial differential equation to predict the evolution of lung tumors during radiotherapy. The evolution of tumor cell density is formulated by three terms: 1) advection describing the advective flux transport of tumor cells, 2) proliferation representing the tumor cell proliferation modeled as Gompertz differential equation, and 3) treatment quantifying the radiotherapeutic efficacy from linear quadratic formulation. We consider that tumor cell density variation can be derived from positron emission tomography images, the novel idea is to model the advection term by calculating 3D optical flow field from sequential images. To estimate patient-specific parameters, we propose an optimization between the predicted and observed images, under a global constraint that the tumor volume decreases exponentially as radiation dose increases. A thresholding on the predicted tumor cell densities is then used to define tumor contours, tumor volumes and maximum standardized uptake values (SUVmax). Results obtained on seven patients show a satisfying agreement between the predicted tumor contours and those drawn by an expert.

MRI of small intrahepatic mass-forming cholangiocarcinoma and atypical small hepatocellular carcinoma (≤ 3 cm) with cirrhosis and chronic viral hepatitis: a comparative study

ObjectiveThe objective was to identify the decision-making magnetic resonance (MR) features in differentiating small intrahepatic mass-forming cholangiocarcinoma (sIMCC) from atypical small hepatocellular carcinoma (sHCC) (≤3 cm) in patients with cirrhosis and chronic viral hepatitis. MethodsSignal features and relative contrast of sHCCs and sIMCCs in T2-weighted and dynamic enhanced imaging were analyzed. A subgroup comparison between the cirrhosis and noncirrhosis chronic viral hepatitis group was also made. ResultsUnivariate analysis revealed that tumor contours (P<.001), signals in T2-weighted (P<.001) and each phase of contrast-enhanced scanning (P<.001), enhancement patterns (P<.001), as well as accompanying findings of tumor capsule (P<.001), hepatic capsule retraction (P<.001), bile duct dilation (P=.031), and transient hepatic intensity difference (P=.002) were different between sIMCC and atypical sHCC. Multivariate analysis indicated that dynamic enhancement patterns (P<.001) and signals in T2-weighted images (P=.024) were independent predictors for differentiation. Confusing MR features were more often observed in the cirrhosis group compared with those in the noncirrhosis chronic viral hepatitis group. ConclusionDynamic enhancement patterns and signals in T2-weighted images were the most important MR features to differentiate sIMCC from atypical sHCC with cirrhosis and chronic viral hepatitis.

Unenhanced CT for the detection of renal cell carcinoma: effect of tumor size and contour type

To evaluate effect of tumor size and contour type for the detection of renal cell carcinoma (RCC) on unenhanced CT. This retrospective institutional review board approved study that includes 111 patients with RCC and 100 patients without RCC who underwent unenhanced CT. Two readers performed a blinded and independent review of the presence of RCC on unenhanced CT. The area under the receiver operating characteristic curves (AUC) was compared by tumor size (<3cm: small, or ≥3cm: large) and contour type (endophytic, mesophytic, or exophytic). For tumor size, the AUC for small RCC (0.70 and 0.78, for reader 1 and reader 2) was significantly lower than that for large RCC (0.97 and 0.99, for reader 1 and reader 2) (p<0.001). As for contour type of tumor, the AUC for endophytic RCC (0.60 and 0.71, for reader 1 and reader 2) was significantly lower than that for mesophytic RCC (0.95 and 0.98, for reader 1 and reader 2) and exophytic RCC (0.98 and 0.99, reader 1 and reader 2) (p<0.001). On unenhanced CT, tumor size and contour type can affect the detection of RCC. While most large or exophytic RCC can be easily detected, the detection of small and endophytic RCC is highly limited.

Brain tumour segmentation in MRI: knowledge-based system and region growing approach

We present in this paper a method for MRI brain tumour segmentation, so we propose a general framework that is a combination of paradigms, in order to have a hybrid segmentation method, automatic and unsupervised. In the first phase, expertise and characteristics derived from MRI images are combined to define heuristics for the development of the classification approach. In the second phase, refinement of the tumour contour is achieved by using the region growing method. The results are good and visually validate by radiologists.

Impact of Perfusion Map Analysis on Early Survival Prediction Accuracy in Glioma Patients

Studies investigating dynamic susceptibility contrast magnetic resonance imaging-determined relative cerebral blood volume (rCBV) maps as a metric of treatment response assessment have generated conflicting results. We evaluated the potential of various analytical techniques to predict survival of patients with glioma treated with chemoradiation. rCBV maps were acquired in patients with high-grade gliomas at 0, 1, and 3 weeks into chemoradiation therapy. Various analytical techniques were applied to the same cohort of serial rCBV data for early assessment of survival. Three different methodologies were investigated: 1) percentage change of whole tumor statistics (i.e., mean, median, and percentiles), 2) physiological segmentation (low rCBV, medium rCBV, or high rCBV), and 3) a voxel-based approach, parametric response mapping (PRM). All analyses were performed using the same tumor contours, which were determined using contrast-enhanced T1-weighted and fluid attenuated inversion recovery images. The predictive potential of each response metric was assessed at 1-year and overall survival. PRM was the only analytical approach found to generate a response metric significantly predictive of patient 1-year survival. Time of acquisition and contour volume were not found to alter the sensitivity of the PRM approach for predicting overall survival. We have demonstrated the importance of the analytical approach in early response assessment using serial rCBV maps. The PRM analysis shows promise as a unified early and robust imaging biomarker of treatment response in patients diagnosed with high-grade gliomas.

The correlation between cell-free DNA and tumour burden was estimated by PET/CT in patients with advanced NSCLC

Background:Cell-free DNA (cfDNA) circulating in the blood holds a possible prognostic value in malignant diseases. Under malignant conditions, the level of cfDNA increases but the biological mechanism remains to be fully understood. We aimed to examine the correlation between cfDNA and total tumour burden defined by positron emission tomography (PET) parameters.Methods:Patients with advanced non-small cell lung cancer (NSCLC) were enrolled into a prospective biomarker trial. Before treatment, plasma was extracted and the level of cfDNA was determined by qPCR. An 18F-fluorodeoxyglucose (18F-FDG) PET/computed tomography (CT) scan was performed and evaluated in terms of metabolic tumour volume (MTV) and total lesion glycolysis (TLG). Tumour contours were delineated semi-automatically by a threshold standardised uptake value (SUV) of 2.5. The primary end point was correlation among cfDNA, MTV and TLG. The secondary end point was overall survival (OS) according to cfDNA, MTV and TLG.Results:Fifty-three patients were included. There were no correlations between cfDNA and MTV (r=0.1) or TLG (r=0.1). cfDNA >75th percentile was correlated with shorter OS (P=0.02), confirmed in a multivariate analysis. MTV>the median was associated with a significantly shorter OS (P=0.02). There was no significant difference in OS according to TLG (P=0.08).Conclusion:Cell-free DNA may not be a simple measure of tumour burden, but seems to reflect more complex mechanisms of tumour biology, making it attractive as an independent prognostic marker.

A multi-step directional generalized gradient vector flow snake for target tumor segmentation in US-guided high-intensity focused ultrasound ablation

Getting precise locations of target tumors can help to ensure ablation of cancerous tissues and avoid unwanted destruction of healthy tissues in high-intensity focused ultrasound (HIFU) treatment system. Because of speckle noise and spurious boundaries in ultrasound images, traditional image segmentation methods are not suitable for achieving the precise locations of target tumors in HIFU ablation. In this paper, a multi-step directional generalized gradient vector flow snake model is introduced for target tumor segmentation. In the first step, the traditional generalized gradient vector flow (GGVF) snake is used to obtain an approximate contour of the tumor. According to the approximate contour, a new distance map is generated. Subsequently, a new directional edge map is created by calculating a scalar product of the gradients of the distance map and the initial image. In this process, the gradient directional information and the magnitude information of the distance map are used to attenuate unwanted edges and highlight the real edges in the new directional edge map. Finally, a refined GGVF field is derived from a diffusion operation of the gradient vectors of the directional edge map. The GGVF field is used to refine the tumor's contour, by directing the approximate contour to edges with the desired gradient directionality. Based on the newly developed snake model, the influences of the spurious boundaries and the speckle noise are significantly reduced in the ultrasound image segmentation. Experimental results indicate that this technique is greatly useful for target tumor segmentation in HIFU treatment system

A cellular automaton model examining the effects of oxygen, hydrogen ions and lactate on early tumour growth

Some tumours are known to exhibit an extracellular pH that is more acidic than the intracellular, creating a 'reversed pH gradient' across the cell membrane and this has been shown to affect their invasive and metastatic potential. Tumour hypoxia also plays an important role in tumour development and has been directly linked to both tumour morphology and aggressiveness. In this paper, we present a hybrid mathematical model of intracellular pH regulation that examines the effect of oxygen and pH on tumour growth and morphology. In particular, we investigate the impact of pH regulatory mechanisms on the cellular pH gradient and tumour morphology. Analysis of the model shows that: low activity of the Na+/H+ exchanger or a high rate of anaerobic glycolysis can give rise to a "fingering" tumour morphology; and a high activity of the lactate/H+ symporter can result in a reversed transmembrane pH gradient across a large portion of the tumour mass. Also, the reversed pH gradient is spatially heterogeneous within the tumour, with a normal pH gradient observed within an intermediate growth layer within the spheroid. We also include a fractal dimension analysis of the simulated tumour contours, in which we compare the fractal dimensions of the simulated tumour surfaces with those found experimentally via photomicrographs.

Variability of tumor area measurements for response assessment in malignant pleural mesothelioma

The measurement of malignant pleural mesothelioma is critical to the assessment of tumor response to therapy. Current response assessment standards utilize summed linear measurements acquired on three computed tomography (CT) sections. The purpose of this study was to evaluate manual area measurements as an alternate response assessment metric, specifically through the study of measurement interobserver variability. Two CT scans from each of 31 patients were collected. Using a computer interface, five observers contoured tumor on three selected CT sections from each baseline scan. Four observers also constructed matched follow-up scan tumor contours for the same 31 patients. Area measurements extracted from these contours were compared using a random effects analysis of variance model to assess relative interobserver variability. The sums of section area measurements were also analyzed, since these area sums are more clinically relevant for response assessment. When each observer's measurements were compared with those of the other four observers, strong correlation was observed. The 95% confidence interval for relative interobserver variability of baseline scan summed area measurements was [-71%, +240%], spanning 311%. For the follow-up scan summed area measurements, the 95% confidence interval for relative interobserver variability was [-41%, +70%], spanning 111%. At both baseline and follow-up, the variability among observers was a significant component of the total variability in both per-section and summed area measurements (p<0.0001). Despite the ability of tumor area measurements to capture tumor burden with greater fidelity than linear tumor thickness measurements, manual area measurements may not be a robust means of response assessment in mesothelioma patients.

PET/CT-based metabolic tumour volume for response prediction of neoadjuvant chemoradiotherapy in oesophageal carcinoma

Neoadjuvant chemoradiotherapy is increasingly used in oesophageal cancer patients. In general, small tumours are associated with a survival benefit compared to large tumours. Little is known, however, about the relationship between initial tumour volume and response to chemoradiotherapy. Therefore, the aim of this study was to determine whether the pretherapy metabolic tumour volume (MTV) on diagnostic PET/CT in oesophageal cancer patients is correlated with response to chemoradiotherapy in the resection specimen. A consecutive series of patients underwent diagnostic PET/CT scanning prior to chemoradiotherapy and oesophagectomy. MTVs were determined on PET/CT and an automated tumour contour was generated using specified standard uptake value thresholds. Response to chemoradiotherapy was determined in the resection specimen according to the scoring system developed by Mandard et al. Patients were divided into different groups according to response to chemoradiotherapy. Between January 2008 and May 2011 a total of 115 patients underwent an oesophagectomy. The MTV determined on diagnostic PET/CT scans was available in 79 patients. Of these 79 patients, 30 (38 %) showed no residual tumour cells at the location of the primary tumour. Three of these patients presented with residual tumour cells in the lymph nodes; 27 patients (34 %) had a complete pathological response. There was a trend towards a better response in patients with a smaller MTV (p = 0.084). This study demonstrated a trend towards a correlation between response to chemoradiotherapy in oesophageal cancer patients and smaller MTVs as determined on diagnostic PET/CT prior to neoadjuvant chemoradiotherapy. However, tumour volumes overlapped between groups, indicating the need for multifactorial parameters as predictors. In addition, a complete local tumour response may be accompanied by residual disease in the regional lymph nodes.

Breast cancer diagnosis based on spiculation feature and neural network techniques

The degree of spiculation of the tumor edge is a particularly relevant indicator of malignancy in the analysis of breast tumoral masses. This paper introduces four new methods for extracting the spiculation feature of a detected breast lesion on mammography by segmenting the contour of the lesion in a number of regions which are separately analysed, determining a characterizing spiculation feature set. In order to differentiate between benign and malignant tumors based on the extracted spiculation sets, an intelligent neural network is first trained on a number of 96 cases of known breast cancer malignancy and then tested for diagnosing and classifying breast cancer tumors. The input of the neural network is thus the extracted spiculation feature set and the output is represented by the histopatological diagnostic given by doctors. Finally, the performance of the introduced methods is analysed depending on the number of regions in which the contour is segmented and the performance-related conclusions are stated for each of the methods.The highlight of this paper is the division of the tumour contour in regions and the assessment of a spiculation indicator for each region, resulting a set of spiculation indicators that characterise the tumour and - by training a neural network - can be used in classifying breast tumours with high performance.

TH‐C‐141‐12: Evaluation of Potential Internal Target Volume for Liver Tumor with Consideration of Respiratory Variation Using Cine MRI

Purpose: Although four‐dimensional CT (4DCT) is valuable for evaluating the respiratory motion and internal target volume (ITV) for liver stereotactic body radiotherapy (SBRT), 4DCT cannot consider the respiratory variation. Here, we developed a novel technique to evaluate the potential ITV with consideration of respiratory variation using cine‐magnetic resonance imaging (cine‐MRI). Methods: We retrospectively evaluated 6 patients who received liver SBRT in 2012. The planning CT (pCT) was acquired with free breathing, and then 4D‐CT and cine‐MRI images were also acquired with the same set‐up condition. The respiratory signal was derived from the peak of the diaphragm on cine‐MRI frames. The motion vectors of 500 feature points in consecutive two frames were calculated using pyramidal implementation of the Lucas‐Kanade optical flow algorithm. The contour of tumor, blood vessel near tumor, or liver was delineated on one frame and the corner points of the contour were shifted onto the other frames using the motion vectors and then the tumor trajectory was calculated. On pCT images, gross tumor volumes delineated on the inspiration and expiration phases of 4D‐CT were three‐dimensionally translated using the tumor trajectory calculated with sagittal and coronal cine‐MRI. The distribution of the existence probability of the tumor was evaluated. Results: The concordance between ITV determined with 4D‐CT and potential ITV were evaluated with Dice similarity coefficient (DSC). When the voxels with the existence probability higher than cut‐off values were considered as potential ITV, the maximum DSC for each patient was 0.88 (range 0.84% to 0.93) and the cut‐off for maximum DSC was 24.5% (range 17% to 32%). Conclusion: We developed a novel technique to evaluate the potential ITV of liver tumor with consideration of the respiratory variation. Potential ITV may be approximately 20% larger than ITV determined with 4DCT. The volume will be larger for patients with large variation in respiration.

Automated tumor size assessment: Consistency of computer measurements with an expert panel.

7566 Background: Manual tumor measurements for use in diagnosis and longitudinal assessment suffer from intra- and inter-observer variability. In practice they are also limited to 1-dimensional diameter measurements because manual contouring of 3D tumor boundaries is impractical. In this work we evaluated a fully automated tumor assessment system in the setting of lung nodules on CT by comparing its tumor size and density measurements against independent measurements made by an expert panel of radiologists. Methods: A new computer-aided detection (CAD) system has been developed that performs fully automated lung nodule detection and measurement. In order to identify the nodule boundary in 3D the system performs automated intensity thresholding, a Euclidean Distance Transformation, and segmentation based on watersheds. The system computes nodule diameter, volume, and mean density in Hounsfield Units (HU). The automated measurements were evaluated against data from the publically available Lung Imaging Database Consortium (LIDC), where each CT scan was reviewed and annotated (with volumetric tumor contours) by an expert panel of four radiologists. Nodule were included from consecutive subjects in the database that were ≥ 4 mm. The CT slice thickness ranged from 0.6 – 3.0 mm. For each nodule measure, the intra-class correlation coefficient (ICC) was computed among the four radiologists and then re-computed with CAD as a fifth observer. Results: 51 lung nodules from 44 subjects were analyzed. The ICCs were computed as shown in the table using a log transformation for volume and diameter. All nodule measures have similar ICCs and overlapping confidence intervals (CI). Conclusions: A new fully automated CAD system has been developed that provides CT lung nodule measurements that are consistent with an expert panel of radiologists. The automated computer system enables practical 3D tumor assessment and reduces measurement variability, which has implications for longitudinal studies. [Table: see text]

18F-fluorodeoxyglucose positron emission tomography/computed tomography and the relationship between fluorodeoxyglucose uptake and the expression of hypoxia-inducible factor-1 , glucose transporter-1 and vascular endothelial growth factor in thymic epithelial tumours

The objective of this study was to evaluate the usefulness of (18)F-fluorodeoxyglucose positron emission tomography/computed tomography (FDG-PET/CT) and the relationships among the expressions of hypoxia-inducible factor-1α (HIF-1α), glucose transporter-1 (Glut-1) and vascular endothelial growth factor (VEGF), histological type, other clinical factors and FDG uptake in thymic epithelial tumours. Thirty-three patients who underwent FDG-PET/CT before treatment were reviewed. All types of tumours were reclassified into three subgroups: low-risk thymomas (types A, AB and B1), high-risk thymomas (types B2 and B3) and thymic carcinomas. Tumour contour, pattern of FDG uptake, tumour size and maximum standardized uptake value (SUVmax) were obtained. Expressions of HIF-1α, Glut-1 and VEGF were analysed immunohistochemically, and these expressions were evaluated using grading scales. FDG uptake was visually recognized in all (100%) tumours. A homogeneous pattern of FDG uptake was increasingly observed in the order of low-risk thymomas to high-risk thymomas to thymic carcinomas (P = 0.016). SUVmax for thymic carcinomas was significantly higher than that for thymomas (P = 0.008). With the optimal cut-off value of SUVmax of 5.6, the sensitivity, specificity and accuracy for diagnosing thymic carcinoma were 0.75, 0.80 and 0.79, respectively. Regarding the mean scoring of HIF-1α, Glut-1 and VEGF, increasing trends were observed in the order of low-risk thymomas to high-risk thymomas to thymic carcinomas. Tumour size revealed a significant correlation with SUVmax (r = 0.60, P < 0.001), and the expression of HIF-1α showed a moderate association, but the expression of Glut-1 showed no correlation with SUVmax. Regarding correlations between the expression of the three markers, there were moderate associations between HIF-1α and Glut-1, and HIF-1α and VEGF, and a significant correlation between Glut-1 and VEGF (r = 0.60, P < 0.001). In type B1 thymoma, HIF-1α and Glut-1 were partly expressed in non-neoplastic immature lymphocytes. FDG-PET/CT should be performed in patients with tumours in the anterior mediastinum because the pattern of FDG uptake and SUVmax are useful in the differential diagnosis of thymic epithelial tumours. Furthermore, the expressions of HIF-1α, Glut-1 and VEGF might be associated with malignancy of thymic epithelial tumours. In contrast, FDG uptake might be dependent on tumour size rather than Glut-1 overexpression.

Three-Dimensional Region-Based Segmentation for Breast Tumors on Sonography

Because malignant and benign breast tumors show different shapes and sizes on sonography, information about tumor shapes and sizes is important for clinical diagnosis. Since sonograms include noise and tissue texture, accurate clinical diagnosis is highly dependent on clinical experience and expertise. However, manually sketching a 3-dimensional (3D) breast tumor contour is a time-consuming and complicated task. Automatic contouring, which provides a contour similar to that of manual sketching of a breast tumor on sonography, may improve diagnostic accuracy. This study presents an efficient method for automatically detecting 3D contours of breast tumors on 3D sonography. The proposed method applies a voxel nearest neighbor filter, a Wiener filter, and an unsharp filter to enhance contrast and reduce noise. After a 3D region-growing algorithm is used to obtain the contour of the breast tumor, postprocessing of the extracted contour is performed to diminish the shadow region of the tumor. This study evaluated 20 tumor cases comprising 10 benign and 10 malignant cases. The results of computer simulation reveal that the proposed 3D segmentation method provides robust contouring for breast sonograms. This approach consistently obtains contours similar to those obtained by manual contouring of a breast tumor and can reduce the time needed to sketch precise contours.

Three-Dimensional Region-Based Segmentation for Breast Tumors on Sonography

Because malignant and benign breast tumors show different shapes and sizes on sonography, information about tumor shapes and sizes is important for clinical diagnosis. Since sonograms include noise and tissue texture, accurate clinical diagnosis is highly dependent on clinical experience and expertise. However, manually sketching a 3-dimensional (3D) breast tumor contour is a time-consuming and complicated task. Automatic contouring, which provides a contour similar to that of manual sketching of a breast tumor on sonography, may improve diagnostic accuracy. This study presents an efficient method for automatically detecting 3D contours of breast tumors on 3D sonography. The proposed method applies a voxel nearest neighbor filter, a Wiener filter, and an unsharp filter to enhance contrast and reduce noise. After a 3D region-growing algorithm is used to obtain the contour of the breast tumor, postprocessing of the extracted contour is performed to diminish the shadow region of the tumor. This study evaluated 20 tumor cases comprising 10 benign and 10 malignant cases. The results of computer simulation reveal that the proposed 3D segmentation method provides robust contouring for breast sonograms. This approach consistently obtains contours similar to those obtained by manual contouring of a breast tumor and can reduce the time needed to sketch precise contours.