Automated Extraction of Cerebral Infarction Region in Head MR Image Using Pseudo Cerebral Infarction Image by CycleGAN

Since recognizing the location and extent of infarction is essential for diagnosis and treatment, many methods using deep learning have been reported. Generally, deep learning requires a large amount of training data. To overcome this problem, we generated pseudo patient images using CycleGAN, which performed image transformation without paired images. Then, we aimed to improve the extraction accuracy by using the generated images for the extraction of cerebral infarction regions. First, we used CycleGAN for data augmentation. Pseudo-cerebral infarction images were generated from healthy images using CycleGAN. Finally, U-Net was used to segment the cerebral infarction region using CycleGAN-generated images. Regarding the extraction accuracy, the Dice index was 0.553 for U-Net with CycleGAN, which was an improvement over U-Net without CycleGAN. Furthermore, the number of false positives per case was 3.75 for U-Net without CycleGAN and 1.23 for U-Net with CycleGAN, respectively. The number of false positives was reduced by approximately 67% by introducing the CycleGAN-generated images to training cases. These results indicate that utilizing CycleGAN-generated images was effective and facilitated the accurate extraction of the infarcted regions while maintaining the detection rate.

Performance of an automated ultrasound device in identifying and tracing the heart in porcine cardiac arrest

Background While intra-arrest echocardiography can be used to guide and monitor chest compression quality, it is not currently feasible on the scene of out-of-hospital cardiac arrests. Rapid and automated sonographic localization of the heart may provide first-responders guidance to an optimal area of compression without requiring them to interpret ultrasound images. In this proof-of-concept porcine study, we sought to describe the performance of an automated ultrasound device in correctly identifying and tracing the borders of the heart in three distinct states: pre-arrest, arrest, and late arrest. Methods An automated ultrasound device (bladder scanner) was placed on the chests of 7 swine, along the left sternal border (4th–8th intercostal spaces). Scanner-generated images were recorded for each space during pre-arrest, arrest, and finally late arrest. 828 images of the LV and LV outflow tract were randomized and 150 (50/state) selected for analysis. Scanner tracings of the heart were then digitally obscured to facilitate tracing by expert reviewers who were blinded to the physiologic state. Reviewer tracings were compared to bladder scanner tracings; with concordance between these images determined via Sørensen–Dice index (SDI). Results When compared to human reviewers, the bladder scanner was able to identify and trace the borders during cardiac arrest. The bladder scanner performed best at the time of arrest (SDI 0.900 ± 0.059). As resuscitation efforts continued and time from initial arrest increased, the scanner’s performance decreased dramatically (SDI 0.597 ± 0.241 in late arrest). Conclusion An automated ultrasound device (bladder scanner) reliably traced porcine hearts during cardiac arrest. It is possible a device could be developed to indicate where compressions should be performed without requiring the operator to interpret ultrasound images. Further investigation into rapid, automated, sonographic localization of the heart to identify the area of compression in out-of-hospital cardiac arrest is warranted.

Investigation of the clinical inter-observer bias in prostate fiducial marker image registration between CT and MR images

Background and purpose Inter-modality image registration between computed tomography (CT) and magnetic resonance (MR) images is associated with systematic uncertainties and the magnitude of these uncertainties is not well documented. The purpose of this study was to investigate the potential uncertainty of gold fiducial marker (GFM) registration for localized prostate cancer and to estimate the inter-observer bias in a clinical setting. Methods Four experienced observers registered CT and MR images for 42 prostate cancer patients. Manual GFM identification was followed by a landmark-based registration. The absolute difference between observers in GFM identification and the displacement of the clinical target volume (CTV) was investigated. The CTV center of mass (CoM) vector displacements, DICE-index and Hausdorff distances for the observer registrations were compared against a clinical baseline registration. The time allocated for the manual registrations was compared. Results Absolute difference in GFM identification between observers ranged from 0.0 to 3.0 mm. The maximum CTV CoM displacement from the clinical baseline was 3.1 mm. Displacements larger than or equal to 1 mm, 2 mm and 3 mm were 46%, 18% and 4%, respectively. No statistically significant difference was detected between observers in terms of CTV displacement. Median DICE-index and Hausdorff distance for the CTV, with their respective ranges were 0.94 [0.70–1.00] and 2.5 mm [0.7–8.7]. Conclusions Registration of CT and MR images using GFMs for localized prostate cancer patients was subject to inter-observer bias on an individual patient level. A CTV displacement as large as 3 mm occurred for individual patients. These results show that GFM registration in a clinical setting is associated with uncertainties, which motivates the removal of inter-modality registrations in the radiotherapy workflow and a transition to an MRI-only workflow for localized prostate cancer.

Deep Learning on MRI Images for Diagnosis of Lung Cancer Spinal Bone Metastasis

This paper aimed to explore the adoption of deep learning algorithms in lung cancer spinal bone metastasis diagnosis. Comprehensive analysis was carried out with the aid of AdaBoost algorithm and Chan-Vese (CV) algorithm. 87 patients with lung cancer spinal bone metastasis were taken as research subjects, and comprehensive evaluation was made in terms of preliminary classification of images, segmentation results, Dice index, and Jaccard coefficient. After the case of misjudgment on whether there was hot spot was excluded, the initial classification accuracy of the AdaBoost algorithm can reach 96.55%. True positive rate (TPR) was 2.3%, and false negative rate (FNR) was 1.15%. 45 MRI images with hot spots were utilized as test set to detect the segmentation accuracy of CV, maximum between-cluster variance method (OTSU), and region growing algorithm. The results showed that the Dice index and Jaccard coefficient of the CV algorithm were 0.8591 and 0.8002, respectively, which were considerably superior to OTSU (0.6125 and 0.5541) and region growing algorithm (0.7293 and 0.6598). In summary, the AdaBoost algorithm was adopted for image preliminary classification, and CV algorithm for image segmentation was ideal for the diagnosis of lung cancer spinal bone metastasis and it was worthy of clinical promotion.

Definition of Local Recurrence Site in Resected Pancreatic Adenocarcinoma: A Multicenter Study (DOLORES-1)

The study aimed to generate a local failure (LF) risk map in resected pancreatic cancer (PC) and validate the results of previous studies, proposing new guidelines for PC postoperative radiotherapy clinical target volume (CTV) delineation. Follow-up computer tomography (CT) of resected PC was retrospectively reviewed by two radiologists identifying LFs and plotting them on a representative patient CT scan. The percentages of LF points randomly extracted based on CTV following the RTOG guidelines and based on the LF database were 70% and 30%, respectively. According to the Kernel density estimation, an LF 3D distribution map was generated and compared with the results of previous studies using a Dice index. Among the 64 resected patients, 59.4% underwent adjuvant treatment. LFs closer to the root of the celiac axis (CA) or the superior mesenteric artery (SMA) were reported in 32.8% and 67.2% cases, respectively. The mean (± standard deviation) distances of LF points to CA and SMA were 21.5 ± 17.9 mm and 21.6 ± 12.1 mm, respectively. The Dice values comparing our iso-level risk maps corresponding to 80% and 90% of the LF probabilistic density and the CTVs-80 and CTVs-90 of previous publications were 0.45–0.53 and 0.58–0.60, respectively. According to the Kernel density approach, a validated LF map was proposed, modeling a new adjuvant CTV based on a PC pattern of failure.

An exploratory study into the influence of laterality and location of hippocampal sclerosis on seizure prognosis and global cortical thinning

In mesial temporal lobe epilepsy (mTLE), the correlation between disease duration, seizure laterality, and rostro-caudal location of hippocampal sclerosis has not been examined in the context of seizure severity and global cortical thinning. In this retrospective study, we analyzed structural 3 T MRI from 35 mTLE subjects. Regions of FLAIR hyperintensity (as an indicator of sclerosis)—based on 2D coronal FLAIR sequences—in the hippocampus were manually segmented, independently and in duplicate; degree of segmentation agreement was confirmed using the DICE index. Segmented lesions were used for separate analyses. First, the correlation of cortical thickness with disease duration and seizure focus laterality was explored using linear model regression. Then, the relationship between the rostro-caudal location of the FLAIR hyperintense signal and seizure severity, based on the Cleveland Clinic seizure freedom score (ccSFS), was explored using probabilistic voxel-wise mapping and functional connectivity analysis from normative data. The mean DICE Index was 0.71 (range 0.60–0.81). A significant correlation between duration of epilepsy and decreased mean whole brain cortical thickness was identified, regardless of seizure laterality (p < 0.05). The slope of cortical volume loss over time, however, was greater in subjects with right seizure focus. Based on probabilistic voxel-wise mapping, FLAIR hyperintensity in the posterior hippocampus was significantly associated with lower ccSFS scores (greater seizure severity). Finally, the right hippocampus was found to have greater brain-wide connectivity, compared to the left side, based on normative connectomic data. We have demonstrated a significant correlation between duration of epilepsy and right-sided seizure focus with global cortical thinning, potentially due to greater brain-wide connectivity. Sclerosis along the posterior hippocampus was associated with greater seizure severity, potentially serving as an important biomarker of seizure outcome after surgery.

Edge-Aware Pyramidal Deformable Network for Unsupervised Registration of Brain MR Images

Deformable image registration is of essential important for clinical diagnosis, treatment planning, and surgical navigation. However, most existing registration solutions require separate rigid alignment before deformable registration, and may not well handle the large deformation circumstances. We propose a novel edge-aware pyramidal deformable network (referred as EPReg) for unsupervised volumetric registration. Specifically, we propose to fully exploit the useful complementary information from the multi-level feature pyramids to predict multi-scale displacement fields. Such coarse-to-fine estimation facilitates the progressive refinement of the predicted registration field, which enables our network to handle large deformations between volumetric data. In addition, we integrate edge information with the original images as dual-inputs, which enhances the texture structures of image content, to impel the proposed network pay extra attention to the edge-aware information for structure alignment. The efficacy of our EPReg was extensively evaluated on three public brain MRI datasets including Mindboggle101, LPBA40, and IXI30. Experiments demonstrate our EPReg consistently outperformed several cutting-edge methods with respect to the metrics of Dice index (DSC), Hausdorff distance (HD), and average symmetric surface distance (ASSD). The proposed EPReg is a general solution for the problem of deformable volumetric registration.

Corticolous Lichens of Belarus. II. Obligate and Facultative Epiphytes

The lichen biota of Belarus lists 406 corticolous species. Of these, 213 (35.7%) species are obligate epiphytes. Crustose lichens make up the majority of obligate epiphytes (157 species, or 73.7%). Apparently, this can indicate the decisive role of the morphology and chemistry of the substrate for the closely contacted lichen thallus. Facultative epiphytes are represented by 193 species, which are quite widely represented by foliose and fruticose life forms (51.3%). Facultative epiphytes inhabit rotting and processed wood, stony substrates, soil (including forest litter), mosses, leaves (needles), root turnouts and metal objects. Wood is inhabited by 154 facultative epiphyte species, of which 80 lichens are strongly epiphyticlignicolous. The lichen diversity of tree bark and wood is relatively similar; the value of the Sørensen-Dice index is equal to 0.51. Soil is the second most important substrate after wood for facultative epiphytes. It is inhabited by 55 lichens, most of which are represented by Cladonia and Peltigera species. 46 species of facultative epiphytes were found on mosses. These are represented mainly by cyanobiont-containing lichens, broad-lobed species, as well as many by the representatives of the genus Cladonia. The stony substrate is suitable for 43 facultative epiphytes species and is characterized by a high specificity of lichen biota. Its Sørensen- Dice index is equal to 0.13. Most of these representatives are common in urban environments. Fungi, lichens, root inversions, leaves, and metal are predominantly inhabited by multisubstrate lichen species.

Multiclass U-Net Segmentation of Brain Electron Microscopy Data

This work focuses on multi-class labeling and segmentation of electron microscopy data. The well-known and state-of-the-art EPFL open dataset has been labeled for 6 classes (instead of 1) and a multi-class version of the U-Net was trained. The new labeled classes are mitochondrion together with its border, mitochondrion’s border (separately), membrane, PSD, axon, vesicle. Our labeling results are available on GitHub. Our study showed that the quality of segmentation is affected by the presence of a sufficient number of specific features that distinguish the selected classes and the representation of these features in the training dataset. With 6-classes segmentation, mitochondria were segmented with the Dice index of 0.94, which is higher than with 5-classes (without mitochondrial boundaries) segmentation (Dice index of 0.892).

RADT-39. 68Ga- DOTATATE CT/PET FOR GAMMA KNIFE RADIOSURGERY PLANNING AND TARGET DELINEATION IN PATIENTS WITH MENINGIOMA

BACKGROUND Stereotactic radiosurgery (SRS) planning for patients with meningiomas can be confounded by difficulty in identifying the tumor boundary, especially in those who have had prior surgery. Recent data have suggested the benefit of 68Ga-DOTATATE CT/PET scans in delineation of meningioma compared to MRI alone. We propose that incorporating 68Ga-DOTATATE PET scans in addition to MRI in SRS planning will provide better target identification and tumor coverage compared to MRI alone. METHODS We reviewed patients with meningioma who had MRI and 68Ga-DOTATATE PET imaging over 12 months. Images were imported into Velocity treatment planning software and separated into two different sessions, one in which only the MRI was accessible, and a second which had the PET scan fused to the MRI. Three different users were asked to contour the residual meningioma as gross tumor volume (GTV) first with MRI alone, and then with the PET/MRI fusion. The volume of each GTV pre-and post-PET fusion was compared and a Dice index was generated. RESULTS Four patients with 6 GTV targets were identified. PET fusion identified new lesions close to the initial GTV targets in 2 patients. The first was a discontinuous dural lesion in the post-op bed. The second was a nodular dural lesion along the left high parietal convexity adjacent to a prior craniectomy and mesh duraplasty site. In the third patient, PET scan identified a greater extent of disease in the skull base. Across all observers, GTV volumes were significantly increased when PET fusion was used. The average volume (cc) increase was 111.6%±66.2%. The average Dice index was 0.58±0.17. CONCLUSION 68Ga-DOTATATE PET scan fused with MRI improved the visualization of meningiomas in patients undergoing SRS. A larger experience is needed to confirm this trend. We have begun to use DOTATATE-PET imaging regularly when imaging patients with meningiomas for SRS.