Quantitative CT perfusion imaging in patients with pancreatic cancer: a systematic review.

Pancreatic ductal adenocarcinoma (PDAC) is the third leading cause of cancer-related death with a 5-year survival rate of 10%. Quantitative CT perfusion (CTP) can provide additional diagnostic information compared to the limited accuracy of the current standard, contrast-enhanced CT (CECT). This systematic review evaluates CTP for diagnosis, grading, and treatment assessment of PDAC. The secondary goal is to provide an overview of scan protocols and perfusion models used for CTP in PDAC. The search strategy combined synonyms for 'CTP' and 'PDAC.' Pubmed, Embase, and Web of Science were systematically searched from January 2000 to December 2020 for studies using CTP to evaluate PDAC. The risk of bias was assessed using QUADAS-2. 607 abstracts were screened, of which 29 were selected for full-text eligibility. 21 studies were included in the final analysis with a total of 760 patients. All studies comparing PDAC with non-tumorous parenchyma found significant CTP-based differences in blood flow (BF) and blood volume (BV). Two studies found significant differences between pathological grades. Two other studies showed that BF could predict neoadjuvant treatment response. A wide variety in kinetic models and acquisition protocol was found among included studies. Quantitative CTP shows a potential benefit in PDAC diagnosis and can serve as a tool for pathological grading and treatment assessment; however, clinical evidence is still limited. To improve clinical use, standardized acquisition and reconstruction parameters are necessary for interchangeability of the perfusion parameters.

Interactive 3D U-net for the segmentation of the pancreas in computed tomography scans.

The increasing incidence of pancreatic cancer will make it the second deadliest cancer in 2030. Imaging based early diagnosis and image guided treatment are emerging potential solutions. Artificial intelligence (AI) can help provide and improve widespread diagnostic expertise and accurate interventional image interpretation. Accurate segmentation of the pancreas is essential to create annotated data sets to train AI, and for computer assisted interventional guidance. Automated deep learning segmentation performance in pancreas computed tomography (CT) imaging is low due to poor grey value contrast and complex anatomy. A good solution seemed a recent interactive deep learning segmentation framework for brain CT that helped strongly improve initial automated segmentation with minimal user input. This method yielded no satisfactory results for pancreas CT, possibly due to a sub-optimal neural network architecture. We hypothesize that a state-of-the-art U-net neural network architecture is better because it can produce a better initial segmentation and is likely to be extended to work in a similar interactive approach. We implemented the existing interactive method, iFCN, and developed an interactive version of U-net method we call iUnet. The iUnet is fully trained to produce the best possible initial segmentation. In interactive mode it is additionally trained on a partial set of layers on user generated scribbles. We compare initial segmentation performance of iFCN and iUnet on a 100CT dataset using dice similarity coefficient analysis. Secondly, we assessed the performance gain in interactive use with three observers on segmentation quality and time. Average automated baseline performance was 78% (iUnet) versus 72% (FCN). Manual and semi-automatic segmentation performance was: 87% in 15 min. for manual, and 86% in 8 min. for iUNet. We conclude that iUnet provides a better baseline than iFCN and can reach expert manual performance significantly faster than manual segmentation in case of pancreas CT. Our novel iUnet architecture is modality and organ agnostic and can be a potential novel solution for semi-automatic medical imaging segmentation in general.

Automatic computer-aided detection of prostate cancer based on multiparametric magnetic resonance image analysis.

In this paper, a fully automatic computer-aided detection (CAD) method is proposed for the detection of prostate cancer. The CAD method consists of multiple sequential steps in order to detect locations that are suspicious for prostate cancer. In the initial stage, a voxel classification is performed using a Hessian-based blob detection algorithm at multiple scales on an apparent diffusion coefficient map. Next, a parametric multi-object segmentation method is applied and the resulting segmentation is used as a mask to restrict the candidate detection to the prostate. The remaining candidates are characterized by performing histogram analysis on multiparametric MR images. The resulting feature set is summarized into a malignancy likelihood by a supervised classifier in a two-stage classification approach. The detection performance for prostate cancer was tested on a screening population of 200 consecutive patients and evaluated using the free response operating characteristic methodology. The results show that the CAD method obtained sensitivities of 0.41, 0.65 and 0.74 at false positive (FP) levels of 1, 3 and 5 per patient, respectively. In conclusion, this study showed that it is feasible to automatically detect prostate cancer at a FP rate lower than systematic biopsy. The CAD method may assist the radiologist to detect prostate cancer locations and could potentially guide biopsy towards the most aggressive part of the tumour.

Automated 3-dimensional segmentation of pelvic lymph nodes in magnetic resonance images.

PURPOSE: Computer aided diagnosis (CAD) of lymph node metastases may help reduce reading time and improve interpretation of the large amount of image data in a 3-D pelvic MRI exam. The purpose of this study was to develop an algorithm for automated segmentation of pelvic lymph nodes from a single seed point, as part of a CAD system for the classification of normal vs metastatic lymph nodes, and to evaluate its performance compared to other algorithms. METHODS: The authors' database consisted of pelvic MR images of 146 consecutive patients, acquired between January 2008 and April 2010. Each dataset included four different MR sequences, acquired after infusion of a lymph node specific contrast medium based on ultrasmall superparamagnetic particles of iron oxide. All data sets were analyzed by two expert readers who, reading in consensus, annotated and manually segmented the lymph nodes. The authors compared four segmentation algorithms: confidence connected region growing (CCRG), extended CCRG (ECC), graph cut segmentation (GCS), and a segmentation method based on a parametric shape and appearance model (PSAM). The methods were ranked based on spatial overlap with the manual segmentations, and based on diagnostic accuracy in a CAD system, with the experts' annotations as reference standard. RESULTS: A total of 2347 manually annotated lymph nodes were included in the analysis, of which 566 contained a metastasis. The mean spatial overlap (Dice similarity coefficient) was: 0.35 (CCRG), 0.57 (ECC), 0.44 (GCS), and 0.46 (PSAM). When combined with the classification system, the area under the ROC curve was: 0.805 (CCRG), 0.890 (ECC), 0.807 (GCS), 0.891 (PSAM), and 0.935 (manual segmentation). CONCLUSIONS: We identified two segmentation methods, ECC and PSAM, that achieve a high diagnostic accuracy when used in conjunction with a CAD system for classification of normal vs metastatic lymph nodes. The manual segmentations still achieve the highest diagnostic accuracy.

Correlation based 3-D segmentation of the left ventricle in pediatric echocardiographic images using radio-frequency data.

Clinical diagnosis of heart disease might be substantially supported by automated segmentation of the endocardial surface in three-dimensional (3-D) echographic images. Because of the poor echogenicity contrast between blood and myocardial tissue in some regions and the inherent speckle noise, automated analysis of these images is challenging. A priori knowledge on the shape of the heart cannot always be relied on, e.g., in children with congenital heart disease, segmentation should be based on the echo features solely. The objective of this study was to investigate the merit of using temporal cross-correlation of radio-frequency (RF) data for automated segmentation of 3-D echocardiographic images. Maximum temporal cross-correlation (MCC) values were determined locally from the RF-data using an iterative 3-D technique. MCC values as well as a combination of MCC values and adaptive filtered, demodulated RF-data were used as an additional, external force in a deformable model approach to segment the endocardial surface and were tested against manually segmented surfaces. Results on 3-D full volume images (Philips, iE33) of 10 healthy children demonstrate that MCC values derived from the RF signal yield a useful parameter to distinguish between blood and myocardium in regions with low echogenicity contrast and incorporation of MCC improves the segmentation results significantly. Further investigation of the MCC over the whole cardiac cycle is required to exploit the full benefit of it for automated segmentation.

In vivo validation of cardiac output assessment in non-standard 3D echocardiographic images.

Automatic segmentation of the endocardial surface in three-dimensional (3D) echocardiographic images is an important tool to assess left ventricular (LV) geometry and cardiac output (CO). The presence of speckle noise as well as the nonisotropic characteristics of the myocardium impose strong demands on the segmentation algorithm. In the analysis of normal heart geometries of standardized (apical) views, it is advantageous to incorporate a priori knowledge about the shape and appearance of the heart. In contrast, when analyzing abnormal heart geometries, for example in children with congenital malformations, this a priori knowledge about the shape and anatomy of the LV might induce erroneous segmentation results. This study describes a fully automated segmentation method for the analysis of non-standard echocardiographic images, without making strong assumptions on the shape and appearance of the heart. The method was validated in vivo in a piglet model. Real-time 3D echocardiographic image sequences of five piglets were acquired in radiofrequency (rf) format. These ECG-gated full volume images were acquired intra-operatively in a non-standard view. Cardiac blood flow was measured simultaneously by an ultrasound transit time flow probe positioned around the common pulmonary artery. Three-dimensional adaptive filtering using the characteristics of speckle was performed on the demodulated rf data to reduce the influence of speckle noise and to optimize the distinction between blood and myocardium. A gradient-based 3D deformable simplex mesh was then used to segment the endocardial surface. A gradient and a speed force were included as external forces of the model. To balance data fitting and mesh regularity, one fixed set of weighting parameters of internal, gradient and speed forces was used for all data sets. End-diastolic and end-systolic volumes were computed from the segmented endocardial surface. The cardiac output derived from this automatic segmentation was validated quantitatively by comparing it with the CO values measured from the volume flow in the pulmonary artery. Relative bias varied between 0 and -17%, where the nominal accuracy of the flow meter is in the order of 10%. Assuming the CO measurements from the flow probe as a gold standard, excellent correlation (r = 0.99) was observed with the CO estimates obtained from image segmentation.

Contrast-enhanced magnetic resonance imaging of the breast: the value of pharmacokinetic parameters derived from fast dynamic imaging during initial enhancement in classifying lesions.

The value of pharmacokinetic parameters derived from fast dynamic imaging during initial enhancement in characterizing breast lesions on magnetic resonance imaging (MRI) was evaluated. Sixty-eight malignant and 34 benign lesions were included. In the scanning protocol, high temporal resolution imaging was combined with high spatial resolution imaging. The high temporal resolution images were recorded every 4.1 s during initial enhancement (fast dynamic analysis). The high spatial resolution images were recorded at a temporal resolution of 86 s (slow dynamic analysis). In the fast dynamic evaluation pharmacokinetic parameters (K(trans), V(e) and k(ep)) were evaluated. In the slow dynamic analysis, each lesion was scored according to the BI-RADS classification. Two readers evaluated all data prospectively. ROC and multivariate analysis were performed. The slow dynamic analysis resulted in an AUC of 0.85 and 0.83, respectively. The fast dynamic analysis resulted in an AUC of 0.83 in both readers. The combination of both the slow and fast dynamic analyses resulted in a significant improvement of diagnostic performance with an AUC of 0.93 and 0.90 (P = 0.02). The increased diagnostic performance found when combining both methods demonstrates the additional value of our method in further improving the diagnostic performance of breast MRI.

Accuracy of rigid CT-FDG-PET image registration of the liver.

Diagnostic and surgical strategies could benefit from accurate localization of liver malignancies via CT-FDG-PET image registration. However, registration uncertainty occurs due to protocol differences in data-acquisition, the limited spatial resolution of positron emission tomography (PET) and the low uptake of 18F-fluorodeoxyglucose (FDG) in normal liver tissue. To assess this uncertainty, methods were presented to estimate registration precision and systematic bias. A semi-automatic, organ-focused method was investigated to minimize the uncertainty well beyond the typical uncertainty of 5-10 mm obtained by commonly available methods. By restricting registration to the liver region and by isolating the liver on computed tomography (CT) from surrounding structures using a thresholding technique, registration was achieved using the mutual information-based method as implemented in insight toolkit (ITK). CT and FDG-PET images of 10 patients with liver metastases were registered rigidly a number of times. Results of the organ-focused method were compared to results of three commonly available methods (a manual, a landmark-based and a 'standard' mutual information-based method) where no dedicated image processing was performed. The proposed method outperformed the other methods with a precision (mean+/-s.d.) of 2.5+/-1.3 mm and a bias of 1.9 mm with a 95% CI of [1.0, 2.8] mm. Unlike the commonly available methods, our approach allows for robust CT-FDG-PET registration of the liver, with an accuracy better than the spatial resolution of the PET scanner that was used.

Post-radiotherapy contrast enhancement changes in fast dynamic MRI of cervical carcinoma.

This pilot study determines fast dynamic gadolinium enhanced MRI contrast enhancement parameters (onset of enhancement and time to peak enhancement) before and after radiotherapy in 10 cervical carcinoma patients. Before radiotherapy, onset of enhancement and time to peak enhancement were early, with a median of 4.5 and 5.2 seconds, respectively. High-grade tumors showed early enhancement, compared with low-grade. After radiotherapy, contrast enhancement patterns differed. In survivors, onset of enhancement after radiotherapy was later than before radiotherapy. In non-survivors, onset of enhancement after radiotherapy was still early. The median difference in onset of enhancement before and after radiotherapy in survivors and non-survivors was an increase of 3.2 and a decrease of 1.1 seconds, respectively. Early onset of enhancement after radiotherapy was a better predictor for survival than a high-signal intensity zone on post radiotherapy unenhanced T1/T2-weighted MRI. It is concluded that enhancement parameters from fast dynamic Gd-enhanced MR images can provide additional functional information with regard to tumor vascularization, and may have prognostic significance. It complements clinical examination and unenhanced MRI in determining the effectiveness of radiotherapy treatment in cervical carcinoma. Future studies will focus on the clinical utility and improvements of the estimation of contrast-enhanced parameters with this new technique.

Accurate estimation of pharmacokinetic contrast-enhanced dynamic MRI parameters of the prostate.

Quantitative analysis of contrast-enhanced dynamic MR images has potential for diagnosing prostate cancer. Contemporary fast acquisition techniques can give sufficiently high temporal resolution to sample the fast dynamics observed in the prostate. Data reduction for parametric visualization requires automatic curve fitting to a pharmacokinetic model, which to date has been performed using least-squares error minimization methods. We observed that these methods often produce unexpectedly noisy estimates, especially for the typically fast, intermediate parameters time-to-peak and start-of-enhancement, resulting in inaccurate pharmacokinetic parameter estimates. We developed a new curve fit method that focuses on the most probable slope. A set of 10 patients annotated using histopathology was used to compare the conventional and new methods. The results show that our new method is significantly more accurate, especially in the relatively less-enhancing peripheral zone. We conclude that estimation accuracy depends on the curve fit method, which is especially important when evaluating the peripheral zone of the prostate.

Fast dynamic gadolinium-enhanced MR imaging of urinary bladder and prostate cancer.

Among the noninvasive imaging modalities, contrast enhanced magnetic resonance (MR) imaging is the most powerful tool with which to visualize vascularity. Common pathology only shows microvessel density, whereas dynamic MR imaging is sensitive to the total endothelial surface area of perfused vessels. Therefore, dynamic MR imaging may be of additional value in tumor staging and in evaluating therapies that affect the perfused microvessel density or surface area, such as chemo-, radiation, or anti-angiogenic therapy. In urinary bladder cancer, this technique results in improved local and nodal staging, in improved separation of transurethral granulation tissue and edema from malignant tumor, and in improved evaluation of the effect of chemotherapy. In prostate cancer, dynamic MR imaging may be of help in problematic cases. This technique can assist in determining seminal vesicle infiltration, in depicting of minimal capsular penetration, and in recognizing tumors within the transitional zone. Also, based on very rapid enhancement, very poorly differentiated tumors can be recognized. Evaluation of the effects of therapy is another promising area, however a lot of research remain to be done. This article reviews some basics of fast enhancement techniques, provides practical information, and shows recent developments, in using these fast techniques for staging and grading of bladder and prostate cancer, and for evaluating the effect of therapy.

Adaptive texture feature extraction with application to ultrasonic image analysis.

Computer texture analysis methods use texture features that are traditionally chosen from a large set of fixed features known in literature. These fixed features are often not specifically designed to the problem at hand, and as a result they may have low discriminative power, and/or may be correlated. Increasing the number of selected fixed features is statistically not a good solution in limited data environments such as medical imaging. For that reason, we developed an adaptive texture feature extraction method (ATFE) that extracts a small number of features that are tuned to the problem at hand. By using a feed-forward neutral network, we ensure that even nonlinear relations are captured from the data. Using extensive, repeated synthetic ultrasonic images, we compared the performance of ATFE with the optimal feature set. We show that the ATFE method is capable of robust operation on small data sets with a performance close to that of the optimal feature set. Another experiment confirms that our ATFE is capable of capturing nonlinear relations from the dataset. We conclude that our method can improve performance in practical, limited dataset situations where an optimal fixed feature set can be hard to find.

Quantitative ultrasonic analysis of liver metastases.

The performance of five features of ultrasonic tissue characterization (UTC) of metastases in vivo in liver was investigated. We acquired serial radiofrequency data sets of 12 patients with metastases in the liver from adenocarcinoma of the colon. Parenchyma and metastases UTC features were estimated in semiautomatically segmented regions. Over 200 metastases were measured in patients and 43 dummy metastases in healthy volunteers. Two attenuation features could be estimated in only 15% of the metastases, and these were not different from those in parenchyma. The texture features signal-to-noise ratio (SNR) could not discriminate real from dummy metastases. Average backscatter intensity, b0, is an established discriminative echographic image feature. However, the metastases that were hypoechoic relative to surrounding parenchyma appeared to be isoechoic relative to normal liver parenchyma. They were visible because of an increased b0 in the surrounding liver parenchyma. Finally, we found an increased backscatter coefficient slope vs. frequency in hypoechoic metastases that may predict a deterioration of lesion contrast at higher transducer frequencies. We conclude that the backscatter coefficient slope can improve detection of metastases, and that b0 measured relative to normal liver parenchyma should be used to correctly correlate metastasis echography with histology.

An in vivo ultrasonic model of liver parenchyma.

Several ultrasonic tissue characterization features are known to discriminate pathological from normal tissue in vivo. Previously, the authors developed an in vivo attenuation- and backscatter estimation method with each frequency dependent coefficient being reduced to a slope and intercept at central frequency. They derived expressions predicting the standard deviation (SD) of these features, assuming a commonly used ultrasonic model of liver parenchyma. In its application to in vivo data, the SD of the intercept features was unexpectedly high. Another feature, signal-to-noise ratio (SNR), showed a significant bias related to the window size. In this paper, the model is extended with the notion of inhomogeneous parenchyma background (IPB). IPB is shown to be present in normal liver parenchyma and is statistically described by a noise term with small amplitude and large correlation cell size. A method is presented to estimate the IPB characteristics. The expressions predicting SD are extended, and an expression is derived predicting the window size bias of the feature SNR. The accuracy and precision estimated from a large in vivo data set shows good agreement with the predictions with the extended model. It is concluded that IPB is a realistic and relevant phenomenon and should be part of the in vivo ultrasonic model of liver parenchyma.

Precision and accuracy of acoustospectrographic parameters.

Theoretical estimates of the standard deviation (STD) of four acoustospectrographic parameters (the intercept and slope of attenuation and backscatter coefficient) are derived. This derivation expands and corrects existing derivations, and is confirmed using simulations based on the adopted theoretical model. A robust parameter estimation method is applied to various phantom measurements, and to in vivo liver scans of healthy human subjects. The measured STD is higher than the theoretically predicted value, and we investigated four possible factors which explain this discrepancy. First, it is shown that the STD and bias after spectrogram calculation are rather insensitive to changes in windowing function, type, length and overlap. Second, we observed that a diffraction correction spectrogram calibrated on a medium different from the one being measured insufficiently corrects the depth-dependency of the parameters, which affects both precision as well as accuracy. We therefore propose a method that constructs an organ-specific diffraction correction spectrogram from the averaged spectrogram of a set of normal organs. We show that the organ-specific correction does not affect STD even in case of previously unseen acquisitions. Third, we introduce local inhomogeneity, which predicts excess STD due to local variations of the physical parameters within an organ (i.e., intrasubject), and global inhomogeneity, which predicts variations between organs (i.e., intersubject). We conclude that our method of estimating STD predicts normal, in vivo data very well, and propose that the deviation from these estimates is a potential tissue characterization parameter.

Fast scan conversion algorithms for displaying ultrasound sector images.

Two fast algorithms for interpolation of ultrasonic sector-scans were developed. Both algorithms are based on line-drawing algorithms and are free from multiplications in the innermost loops. The algorithms were compared to the following conventional interpolators: 2-D windowed sinc, bicubic spline, 4 x 4 point bicubic spline, bilinear, and nearest neighbor. The most accurate of the two new algorithms is about eight times faster than nearest neighbor interpolation. The quantitative errors are of the same order as the errors of the nearest neighbor interpolator. The subjective image quality is between nearest neighbor and bilinear interpolation.