Multi-modal trained artificial intelligence solution to triage chest X-ray for COVID-19 using pristine ground-truth, versus radiologists.

The front-line imaging modalities computed tomography (CT) and X-ray play important roles for triaging COVID patients. Thoracic CT has been accepted to have higher sensitivity than a chest X-ray for COVID diagnosis. Considering the limited access to resources (both hardware and trained personnel) and issues related to decontamination, CT may not be ideal for triaging suspected subjects. Artificial intelligence (AI) assisted X-ray based application for triaging and monitoring require experienced radiologists to identify COVID patients in a timely manner with the additional ability to delineate and quantify the disease region is seen as a promising solution for widespread clinical use. Our proposed solution differs from existing solutions presented by industry and academic communities. We demonstrate a functional AI model to triage by classifying and segmenting a single chest X-ray image, while the AI model is trained using both X-ray and CT data. We report on how such a multi-modal training process improves the solution compared to single modality (X-ray only) training. The multi-modal solution increases the AUC (area under the receiver operating characteristic curve) from 0.89 to 0.93 for a binary classification between COVID-19 and non-COVID-19 cases. It also positively impacts the Dice coefficient (0.59 to 0.62) for localizing the COVID-19 pathology. To compare the performance of experienced readers to the AI model, a reader study is also conducted. The AI model showed good consistency with respect to radiologists. The DICE score between two radiologists on the COVID group was 0.53 while the AI had a DICE value of 0.52 and 0.55 when compared to the segmentation done by the two radiologists separately. From a classification perspective, the AUCs of two readers was 0.87 and 0.81 while the AUC of the AI is 0.93 based on the reader study dataset. We also conducted a generalization study by comparing our method to the-state-art methods on independent datasets. The results show better performance from the proposed method. Leveraging multi-modal information for the development benefits the single-modal inferencing.

Magnetic Resonance Imaging-Based Delineation of Organs at Risk in the Head and Neck Region.

Purpose: The aim of this article is to establish a comprehensive contouring guideline for treatment planning using only magnetic resonance images through an up-to-date set of organs at risk (OARs), recommended organ boundaries, and relevant suggestions for the magnetic resonance imaging (MRI)-based delineation of OARs in the head and neck (H&N) region. Methods and Materials: After a detailed review of the literature, MRI data were collected from the H&N region of healthy volunteers. OARs were delineated in the axial, coronal, and sagittal planes on T2-weighted sequences. Every contour defined was revised by 4 radiation oncologists and subsequently by 2 independent senior experts (H&N radiation oncologist and radiologist). After revision, the final structures were presented to the consortium partners. Results: A definitive consensus was reached after multi-institutional review. On that basis, we provided a detailed anatomic and functional description and specific MRI characteristics of the OARs. Conclusions: In the era of precision radiation therapy, the need for well-built, straightforward contouring guidelines is on the rise. Precise, uniform, delineation-based, automated OAR segmentation on MRI may lead to increased accuracy in terms of organ boundaries and analysis of dose-dependent sequelae for an adequate definition of normal tissue complication probability.

Automated liver volumetry in orthotopic liver transplantation using multiphase acquisitions on MDCT.

OBJECTIVE: The aim of this study was to test a new automated hepatic volumetry technique by comparing the accuracies and postprocessing times of manual and automated liver volume segmentation methods in a patient population undergoing orthotopic liver transplantation so that liver volume could be determined on pathology as the standard of reference. CONCLUSION: Both manual and automated multiphase MDCT-based volume measurements were strongly correlated to liver volume (Pearson correlation coefficient, r = 0.87 [p < 0.0001] and 0.90 [p < 0.0001], respectively). Automated multiphase segmentation was significantly more rapid than manual segmentation (mean time, 16 ± 5 [SD] and 86 ± 3 seconds, respectively; p = 0.01). Overall, automated liver volumetry based on multiphase CT acquisitions is feasible and more rapid than manual segmentation.

Subventricular zone volumetric and dosimetric changes during postoperative brain tumor irradiation and its impact on overall survival.

PURPOSE: The aim of this retrospective study was to investigate the relationship between the dose to the subventricular zone (SVZ) and overall survival (OS) of 41 patients with glioblastoma multiforme (GBM), who were treated with an adaptive approach involving repeated topometric CT and replanning at two-thirds (40 Gy) of their course of postoperative radiotherapy for planning of a 20 Gy boost. METHODS: We examined changes in the ipsilateral lateral ventricle (LV) and SVZ (iLV and iSVZ), as well as in the contralateral LV and SVZ (cLV and cSVZ). We evaluated the volumetric changes on both planning CT scans (primary CT1 and secondary CT2). The survival of the GBM patients was analyzed using the Kaplan-Meier method; the multivariate Cox regression was also performed. RESULTS: Median follow-up and OS were 34.5 months and 17.6 months, respectively. LV and SVZ structures exhibited significant volumetric changes on CT2, resulting in an increase of dose coverage. At a cut-off point of 58 Gy, a significant correlation was detected between the iSVZ2 mean dose and OS (27.8 vs 15.6 months, p = 0.048). In a multivariate analysis, GBM patients with a shorter time to postoperative chemoradiotherapy (<3.8 weeks), with good performance status (≥70%) and higher mean dose (≥58 Gy) to the iSVZ2 had significantly better OS. CONCLUSIONS: Significant anatomical and dose distribution changes to the brain structures were observed, which have a relevant impact on the dose-effect relationship for GBM; therefore, involving the iSVZ in the target volume should be considered and adapted to the changes.

Automatic recognition of anatomical regions in three-dimensional medical images.

This paper presents a method that detects anatomy regions in three-dimensional medical images. The method labels each axial slice of the image according to the anatomy region it belongs to. The detected regions are the head (and neck), the chest, the abdomen, the pelvis, and the legs. The proposed method consists of two main parts. The core of the algorithm is based on a two-dimensional feature extraction that is followed by a random forest classification. This recognition process achieves an overall accuracy of 91.5% in slice classification, but it cannot always provide fully consistent labeling. The subsequent post-processing step incorporates the expected sequence and size of the human anatomy regions in order to improve the accuracy of the labeling. In this part of the algorithm the detected anatomy regions (represented by Gaussian distributions) are fitted to the region probabilities provided by the random forest classifier. The proposed method was evaluated on a set of whole-body MR images. The results demonstrate that the accuracy of the labeling can be increased to 94.1% using the presented post-processing. In order to demonstrate the robustness of the proposed method it was applied to partial MRI scans of different sizes (cut from the whole-body examinations). According to the results the proposed method works reliably (91.3%) for partial body scans (having as little length as 35cm) as well.

Automated liver lesion detection in CT images based on multi-level geometric features.

PURPOSE: Due to the increasing number of liver cancer cases in clinical practice, there is a significant need for efficient tools for computer-assisted liver lesion analysis. A wide range of clinical applications, such as lesion characterization, quantification and follow-up, can be facilitated by automated liver lesion detection. Liver lesions vary significantly in size, shape, density and heterogeneity, which make them difficult to detect automatically. The goal of this work was to develop a method that can detect all types of liver lesions with high sensitivity and low false positive rate within a short run time. METHODS: The proposed method identifies abnormal regions in liver CT images based on their intensity using a multi-level segmentation approach. The abnormal regions are analyzed from the inside-out using basic geometric features (such as asymmetry, compactness or volume). Using this multi-level shape characterization, the abnormal regions are classified into lesions and other region types (including vessel, liver boundary). The proposed analysis also allows defining the contour of each finding. The method was trained on a set of 55 cases involving 120 lesions and evaluated on a set of 30 images involving 59 (various types of) lesions, which were manually contoured by a physician. RESULTS: The proposed algorithm demonstrated a high detection rate (92 %) at a low (1.7) false positive per case (precision 51 %), when the method was started from a manually contoured liver. The same level of false positive per case (1.6) and precision (51 %) was achieved at a somewhat lower detection rate (85 %), when the volume of interest was defined by a fully automated liver segmentation. CONCLUSIONS: The proposed method can efficiently detect liver lesions irrespective of their size, shape, density and heterogeneity within half a minute. According to the evaluation, its accuracy is competitive with the actual state-of-the-art approaches, and the contour of the detected findings is acceptable in most of the cases. Future work shall focus on more precise lesion contouring so that the proposed method can be a solid basis for fully automated liver tumour burden estimation.

Virtual volume resection using multi-resolution triangular representation of B-spline surfaces.

Computer assisted analysis of organs has an important role in clinical diagnosis and therapy planning. As well as the visualization, the manipulation of 3-dimensional (3D) objects are key features of medical image processing tools. The goal of this work was to develop an efficient and easy to use tool that allows the physician to partition a segmented organ into its segments or lobes. The proposed tool allows the user to define a cutting surface by drawing some traces on 2D sections of a 3D object, cut the object into two pieces with a smooth surface that fits the input traces, and iterate the process until the object is partitioned at the desired level. The tool is based on an algorithm that interpolates the user-defined traces with B-spline surface and computes a binary cutting volume that represents the different sides of the surface. The computation of the cutting volume is based on the multi-resolution triangulation of the B-spline surface. The proposed algorithm was integrated into an open-source medical image processing framework. Using the tool, the user can select the object to be partitioned (e.g. segmented liver), define the cutting surface based on the corresponding medical image (medical image visualizing the internal structure of the liver), cut the selected object, and iterate the process. In case of liver segment separation, the cuts can be performed according to a predefined sequence, which makes it possible to label the temporary as well as the final partitions (lobes, segments) automatically. The presented tool was evaluated for anatomical segment separation of the liver involving 14 cases and virtual liver tumor resection involving one case. The segment separation was repeated 3 different times by one physician for all cases, and the average and the standard deviation of segment volumes were computed. According to the test experiences the presented algorithm proved to be efficient and user-friendly enough to perform free form cuts for liver segment separation and virtual liver tumor resection. The volume quantification of segments showed good correlation with the prior art and the vessel-based liver segment separation, which demonstrate the clinical usability of the presented method.

Liver segmentation for contrast-enhanced MR images using partitioned probabilistic model.

PURPOSE: Liver volume segmentation is important in computer assisted diagnosis and therapy planning of liver tumors. Manual segmentation is time-consuming, tedious and error prone, so automated methods are needed. Automatic segmentation of MR images is more challenging than for CT images, so a robust system was developed. METHODS: An intensity-based segmentation method that uses probabilistic model to increase the precision of the segmentation was developed. The model was build based on 60 manually contoured liver CT exams and partitioned into 8 parts according to the (Couinaud) segmental anatomy of the liver. The partitioning allows using different intensity statistics in different parts of the organ, which makes it insensitive to local intensity differences from MR artifacts or pathology. The method employs a modality independent model with registration that exploits some LAVA image characteristics. This dependence can be eliminated to adapt the segmentation method for a wide range of MR images. RESULTS: The method was evaluated using eight representative, manually segmented MR LAVA exams. The results show that the method can accurately segment the liver volume despite various MR artifacts and pathology. The evaluation shows that the proposed method provides more precise segmentation (6% average absolute relative volume error) compared with global intensity statistics for the whole organ (20% average absolute relative volume error). The compute time of the method was 30 s in average, which is acceptable for wide range of clinical applications. CONCLUSION: An automatic method that can segment the liver in contrast-enhanced MR LAVA images was developed and tested. The results demonstrate that the method is feasible, efficient and robust to artifacts and pathology.

Automatic segmentation of the liver from multi- and single-phase contrast-enhanced CT images.

Segmentation of contrast-enhanced abdominal CT images is required by many clinical applications of computer aided diagnosis and therapy planning. The research on automated methods involves different organs among which the liver is the most emphasized. In the current clinical practice more images (at different phases) are acquired from the region of interest in case of a contrast-enhanced abdominal CT examination. The majority of the existing methods, however, use only the portal-venous image to segment the liver. This paper presents a method that automatically segments the liver by combining more phases of the contrast-enhanced CT examination. The method uses region-growing facilitated by pre- and post-processing functions, which incorporate anatomical and multi-phase information to eliminate over- and under-segmentation. Another method, which uses only the portal-venous phase to segment the liver automatically, is also presented. Both methods were evaluated using different datasets, which showed that the result of multi-phase method can be used without or after minor correction in nearly 94% of the cases, and the single-phase method can provide result comparable with non-expert manual segmentation in 90% of the cases. The comparison of the two methods demonstrates that automatic segmentation is more reliable when the information of more phases is combined.

Comparison and evaluation of methods for liver segmentation from CT datasets.

This paper presents a comparison study between 10 automatic and six interactive methods for liver segmentation from contrast-enhanced CT images. It is based on results from the "MICCAI 2007 Grand Challenge" workshop, where 16 teams evaluated their algorithms on a common database. A collection of 20 clinical images with reference segmentations was provided to train and tune algorithms in advance. Participants were also allowed to use additional proprietary training data for that purpose. All teams then had to apply their methods to 10 test datasets and submit the obtained results. Employed algorithms include statistical shape models, atlas registration, level-sets, graph-cuts and rule-based systems. All results were compared to reference segmentations five error measures that highlight different aspects of segmentation accuracy. All measures were combined according to a specific scoring system relating the obtained values to human expert variability. In general, interactive methods reached higher average scores than automatic approaches and featured a better consistency of segmentation quality. However, the best automatic methods (mainly based on statistical shape models with some additional free deformation) could compete well on the majority of test images. The study provides an insight in performance of different segmentation approaches under real-world conditions and highlights achievements and limitations of current image analysis techniques.