Panagiotis Korfiatis

Machine Learning for Medical Imaging

Machine learning is a technique for recognizing patterns that can be applied to medical images. Although it is a powerful tool that can help in rendering medical diagnoses, it can be misapplied. Machine learning typically begins with the machine learning algorithm system computing the image features that are believed to be of importance in making the prediction or diagnosis of interest. The machine learning algorithm system then identifies the best combination of these image features for classifying the image or computing some metric for the given image region. There are several methods that can be used, each with different strengths and weaknesses. There are open-source versions of most of these machine learning methods that make them easy to try and apply to images. Several metrics for measuring the performance of an algorithm exist; however, one must be aware of the possible associated pitfalls that can result in misleading metrics. More recently, deep learning has started to be used; this method has the benefit that it does not require image feature identification and calculation as a first step; rather, features are identified as part of the learning process. Machine learning has been used in medical imaging and will have a greater influence in the future. Those working in medical imaging must be aware of how machine learning works. ©RSNA, 2017.

Automatic total kidney volume measurement on follow-up magnetic resonance images to facilitate monitoring of autosomal dominant polycystic kidney disease progression

BACKGROUND Renal imaging examinations provide high-resolution information about the anatomic structure of the kidneys and are used to measure total kidney volume (TKV) in autosomal dominant polycystic kidney disease (ADPKD) patients. TKV has become the gold-standard image biomarker for ADPKD progression at early stages of the disease and is used in clinical trials to characterize treatment efficacy. Automated methods to segment the kidneys and measure TKV are desirable because of the long time requirement for manual approaches such as stereology or planimetry tracings. However, ADPKD kidney segmentation is complicated by a number of factors, including irregular kidney shapes and variable tissue signal at the kidney borders. METHODS We describe an image processing approach that overcomes these problems by using a baseline segmentation initialization to provide automatic segmentation of follow-up scans obtained years apart. We validated our approach using 20 patients with complete baseline and follow-up T1-weighted magnetic resonance images. Both manual tracing and stereology were used to calculate TKV, with two observers performing manual tracings and one observer performing repeat tracings. Linear correlation and Bland-Altman analysis were performed to compare the different approaches. RESULTS Our automated approach measured TKV at a level of accuracy (mean difference ± standard error = 0.99 ± 0.79%) on par with both intraobserver (0.77 ± 0.46%) and interobserver variability (1.34 ± 0.70%) of manual tracings. All approaches had excellent agreement and compared favorably with ground-truth manual tracing with interobserver, stereological and automated approaches having 95% confidence intervals ∼ ± 100 mL. CONCLUSIONS Our method enables fast, cost-effective and reproducible quantification of ADPKD progression that will facilitate and lower the costs of clinical trials in ADPKD and other disorders requiring accurate, longitudinal kidney quantification. In addition, it will hasten the routine use of TKV as a prognostic biomarker in ADPKD.

Impact of Software Modeling on the Accuracy of Perfusion MRI in Glioma

BACKGROUND AND PURPOSE: Relative cerebral blood volume, as measured by T2*-weighted dynamic susceptibility-weighted contrast-enhanced MRI, represents the most robust and widely used perfusion MR imaging metric in neuro-oncology. Our aim was to determine whether differences in modeling implementation will impact the correction of leakage effects (from blood-brain barrier disruption) and the accuracy of relative CBV calculations as measured on T2*-weighted dynamic susceptibility-weighted contrast-enhanced MR imaging at 3T field strength. MATERIALS AND METHODS: This study included 52 patients with glioma undergoing DSC MR imaging. Thirty-six patients underwent both non-preload dose– and preload dose–corrected DSC acquisitions, with 16 patients undergoing preload dose–corrected acquisitions only. For each acquisition, we generated 2 sets of relative CBV metrics by using 2 separate, widely published, FDA-approved commercial software packages: IB Neuro and nordicICE. We calculated 4 relative CBV metrics within tumor volumes: mean relative CBV, mode relative CBV, percentage of voxels with relative CBV > 1.75, and percentage of voxels with relative CBV > 1.0 (fractional tumor burden). We determined Pearson (r) and Spearman (ρ) correlations between non-preload dose– and preload dose–corrected metrics. In a subset of patients with recurrent glioblastoma (n = 25), we determined receiver operating characteristic area under the curve for fractional tumor burden accuracy to predict the tissue diagnosis of tumor recurrence versus posttreatment effect. We also determined correlations between rCBV and microvessel area from stereotactic biopsies (n = 29) in 12 patients. RESULTS: With IB Neuro, relative CBV metrics correlated highly between non-preload dose– and preload dose–corrected conditions for fractional tumor burden (r = 0.96, ρ = 0.94), percentage > 1.75 (r = 0.93, ρ = 0.91), mean (r = 0.87, ρ = 0.86), and mode (r = 0.78, ρ = 0.76). These correlations dropped substantially with nordicICE. With fractional tumor burden, IB Neuro was more accurate than nordicICE in diagnosing tumor versus posttreatment effect (area under the curve = 0.85 versus 0.67) (P < .01). The highest relative CBV–microvessel area correlations required preload dose and IB Neuro (r = 0.64, ρ = 0.58, P = .001). CONCLUSIONS: Different implementations of perfusion MR imaging software modeling can impact the accuracy of leakage correction, relative CBV calculation, and correlations with histologic benchmarks.

Toolkits and Libraries for Deep Learning

Deep learning is an important new area of machine learning which encompasses a wide range of neural network architectures designed to complete various tasks. In the medical imaging domain, example tasks include organ segmentation, lesion detection, and tumor classification. The most popular network architecture for deep learning for images is the convolutional neural network (CNN). Whereas traditional machine learning requires determination and calculation of features from which the algorithm learns, deep learning approaches learn the important features as well as the proper weighting of those features to make predictions for new data. In this paper, we will describe some of the libraries and tools that are available to aid in the construction and efficient execution of deep learning as applied to medical images.

Utilizing magnetization transfer imaging to investigate tissue remodeling in a murine model of autosomal dominant polycystic kidney disease.

Noninvasive imaging techniques that quantify renal tissue composition are needed to more accurately ascertain prognosis and monitor disease progression in polycystic kidney disease (PKD). Given the success of magnetization transfer (MT) imaging to characterize various tissue remodeling pathologies, it was tested on a murine model of autosomal dominant PKD.

Semi-automated segmentation of pre-operative low grade gliomas in magnetic resonance imaging.

BackgroundSegmentation of pre-operative low-grade gliomas (LGGs) from magnetic resonance imaging is a crucial step for studying imaging biomarkers. However, segmentation of LGGs is particularly challenging because they rarely enhance after gadolinium administration. Like other gliomas, they have irregular tumor shape, heterogeneous composition, ill-defined tumor boundaries, and limited number of image types. To overcome these challenges we propose a semi-automated segmentation method that relies only on T2-weighted (T2W) and optionally post-contrast T1-weighted (T1W) images.MethodsFirst, the user draws a region-of-interest (ROI) that completely encloses the tumor and some normal tissue. Second, a normal brain atlas and post-contrast T1W images are registered to T2W images. Third, the posterior probability of each pixel/voxel belonging to normal and abnormal tissues is calculated based on information derived from the atlas and ROI. Finally, geodesic active contours use the probability map of the tumor to shrink the ROI until optimal tumor boundaries are found. This method was validated against the true segmentation (TS) of 30 LGG patients for both 2D (1 slice) and 3D. The TS was obtained from manual segmentations of three experts using the Simultaneous Truth and Performance Level Estimation (STAPLE) software. Dice and Jaccard indices and other descriptive statistics were computed for the proposed method, as well as the experts’ segmentation versus the TS. We also tested the method with the BraTS datasets, which supply expert segmentations.Results and discussionFor 2D segmentation vs. TS, the mean Dice index was 0.90 ± 0.06 (standard deviation), sensitivity was 0.92, and specificity was 0.99. For 3D segmentation vs. TS, the mean Dice index was 0.89 ± 0.06, sensitivity was 0.91, and specificity was 0.99. The automated results are comparable with the experts’ manual segmentation results.ConclusionsWe present an accurate, robust, efficient, and reproducible segmentation method for pre-operative LGGs.

Semiautomated Segmentation of Polycystic Kidneys in T2-Weighted MR Images

OBJECTIVE The objective of the present study is to develop and validate a fast, accurate, and reproducible method that will increase and improve institutional measurement of total kidney volume and thereby avoid the higher costs, increased operator processing time, and inherent subjectivity associated with manual contour tracing. MATERIALS AND METHODS We developed a semiautomated segmentation approach, known as the minimal interaction rapid organ segmentation (MIROS) method, which results in human interaction during measurement of total kidney volume on MR images being reduced to a few minutes. This software tool automatically steps through slices and requires rough definition of kidney boundaries supplied by the user. The approach was verified on T2-weighted MR images of 40 patients with autosomal dominant polycystic kidney disease of varying degrees of severity. RESULTS The MIROS approach required less than 5 minutes of user interaction in all cases. When compared with the ground-truth reference standard, MIROS showed no significant bias and had low variability (mean ± 2 SD, 0.19% ± 6.96%). CONCLUSION The MIROS method will greatly facilitate future research studies in which accurate and reproducible measurements of cystic organ volumes are needed.

Variability and accuracy of different software packages for dynamic susceptibility contrast magnetic resonance imaging for distinguishing glioblastoma progression from pseudoprogression

Abstract. Determining whether glioblastoma multiforme (GBM) is progressing despite treatment is challenging due to the pseudoprogression phenomenon seen on conventional MRIs, but relative cerebral blood volume (CBV) has been shown to be helpful. As CBV’s calculation from perfusion-weighted images is not standardized, we investigated whether there were differences between three FDA-cleared software packages in their CBV output values and subsequent performance regarding predicting survival/progression. Forty-five postradiation therapy GBM cases were retrospectively identified as having indeterminate MRI findings of progression versus pseudoprogression. The dynamic susceptibility contrast MR images were processed with different software and three different relative CBV metrics based on the abnormally enhancing regions were computed. The intersoftware intraclass correlation coefficients were 0.8 and below, depending on the metric used. No statistically significant difference in progression determination performance was found between the software packages, but performance was better for the cohort imaged at 3.0 T versus those imaged at 1.5 T for many relative CBV metric and classification criteria combinations. The results revealed clinically significant variation in relative CBV measures based on the software used, but minimal interoperator variation. We recommend against using specific relative CBV measurement thresholds for GBM progression determination unless the same software or processing algorithm is used.

Image texture features predict renal function decline in patients with autosomal dominant polycystic kidney disease

Magnetic resonance imaging (MRI) examinations provide high-resolution information about the anatomic structure of the kidneys and are used to measure total kidney volume (TKV) in patients with Autosomal Dominant Polycystic Kidney Disease (ADPKD). Height-adjusted TKV (HtTKV) has become the gold-standard imaging biomarker for ADPKD progression at early stages of the disease when estimated glomerular filtration rate (eGFR) is still normal. However, HtTKV does not take advantage of the wealth of information provided by MRI. Here we tested whether image texture features provide additional insights into the ADPKD kidney that may be used as complementary information to existing biomarkers. A retrospective cohort of 122 patients from the Consortium for Radiologic Imaging Studies of Polycystic Kidney Disease (CRISP) study was identified who had T2-weighted MRIs and eGFR values over 70 mL/min/1.73m2 at the time of their baseline scan. We computed nine distinct image texture features for each patient. The ability of each feature to predict subsequent progression to CKD stage 3A, 3B, and 30% reduction in eGFR at eight-year follow-up was assessed. A multiple linear regression model was developed incorporating age, baseline eGFR, HtTKV, and three image texture features identified by stability feature selection (Entropy, Correlation, and Energy). Including texture in a multiple linear regression model (predicting percent change in eGFR) improved Pearson correlation coefficient from -0.51 (using age, eGFR, and HtTKV) to -0.70 (adding texture). Thus, texture analysis offers an approach to refine ADPKD prognosis and should be further explored for its utility in individualized clinical decision making and outcome prediction.

MIRMAID: A Content Management System for Medical Image Analysis Research.

Today, a typical clinical study can involve thousands of participants, with imaging data acquired over several time points across multiple institutions. The additional associated information (metadata) accompanying these data can cause data management to be a study-hindering bottleneck. Consistent data management is crucial for large-scale modern clinical imaging research studies. If the study is to be used for regulatory submissions, such systems must be able to meet regulatory compliance requirements for systems that manage clinical image trials, including protecting patient privacy. Our aim was to develop a system to address these needs by leveraging the capabilities of an open-source content management system (CMS) that has a highly configurable workflow; has a single interface that can store, manage, and retrieve imaging-based studies; and can handle the requirement for data auditing and project management. We developed a Web-accessible CMS for medical images called Medical Imaging Research Management and Associated Information Database (MIRMAID). From its inception, MIRMAID was developed to be highly flexible and to meet the needs of diverse studies. It fulfills the need for a complete system for medical imaging research management.