Clinical validation of the use of prototype software for automatic cartilage segmentation to quantify knee cartilage in volunteers

Background The cartilage segmentation algorithms make it possible to accurately evaluate the morphology and degeneration of cartilage. There are some factors (location of cartilage subregions, hydrarthrosis and cartilage degeneration) that may influence the accuracy of segmentation. It is valuable to evaluate and compare the accuracy and clinical value of volume and mean T2* values generated directly from automatic knee cartilage segmentation with those from manually corrected results using prototype software. Method Thirty-two volunteers were recruited, all of whom underwent right knee magnetic resonance imaging examinations. Morphological images were obtained using a three-dimensional (3D) high-resolution Double-Echo in Steady-State (DESS) sequence, and biochemical images were obtained using a two-dimensional T2* mapping sequence. Cartilage score criteria ranged from 0 to 2 and were obtained using the Whole-Organ Magnetic Resonance Imaging Score (WORMS). The femoral, patellar, and tibial cartilages were automatically segmented and divided into subregions using the post-processing prototype software. Afterwards, all the subregions were carefully checked and manual corrections were done where needed. The dice coefficient correlations for each subregion by the automatic segmentation were calculated. Results Cartilage volume after applying the manual correction was significantly lower than automatic segmentation (P < 0.05). The percentages of the cartilage volume change for each subregion after manual correction were all smaller than 5%. In all the subregions, the mean T2* relaxation time within manual corrected subregions was significantly lower than in regions after automatic segmentation (P < 0.05). The average time for the automatic segmentation of the whole knee was around 6 min, while the average time for manual correction of the whole knee was around 27 min. Conclusions Automatic segmentation of cartilage volume has a high dice coefficient correlation and it can provide accurate quantitative information about cartilage efficiently without individual bias. Advances in knowledge: Magnetic resonance imaging is the most promising method to detect structural changes in cartilage tissue. Unfortunately, due to the structure and morphology of the cartilages obtaining accurate segmentations can be problematic. There are some factors (location of cartilage subregions, hydrarthrosis and cartilage degeneration) that may influence segmentation accuracy. We therefore assessed the factors that influence segmentations error.

Segmentation Errors in the Measurement of Volumetric Parameters by Swept-Source Anterior Segment Optical Coherence Tomography

Purpose: To investigate the error rate of segmentation in the automatic measurement of anterior chamber volume (ACV) and iris volume (IV) by swept-source anterior segment optical coherence tomography (SS-ASOCT) in narrow-angle and wide-angle eyes.Methods: In this study, fifty eyes from 25 narrow-angle subjects and fifty eyes from 25 wide-angle subjects were enrolled. SS-ASOCT examinations were performed and each SS-ASOCT scan was reviewed, and segmentation errors in the automatic measurement of ACV and IV were classified and manually corrected. Error rates were compared between the narrow-angle and the wide-angle groups, and ACV and IV before and after manual correction were compared.Results: A total of 12,800 SS-ASOCT scans were reviewed. Segmentation error rates of angle recess, iris root, posterior surface of the iris, pupil margin, and anterior surface of the lens were 84.06, 93.30, 13.15, 59.21, and 25.27%, respectively. Segmentation errors of angle recess, iris root, posterior surface of the iris, and pupil margin occurred more frequently in narrow-angle eyes, while more segmentation errors of the anterior surface of the lens were found in wide-angle eyes (all P &lt; 0.001). ACV decreased and IV increased significantly after manual correction of segmentation errors in both groups (all P &lt; 0.01).Conclusion: Segmentation errors were prevalent in the volumetric measurement by SS-ASOCT, particularly in narrow-angle eyes, leading to mismeasurement of ACV and IV.

Creating a training set for artificial intelligence from initial segmentations of airways

Airways segmentation is important for research about pulmonary disease but require a large amount of time by trained specialists. We used an openly available software to improve airways segmentations obtained from an artificial intelligence (AI) tool and retrained the tool to get a better performance. Fifteen initial airway segmentations from low-dose chest computed tomography scans were obtained with a 3D-Unet AI tool previously trained on Danish Lung Cancer Screening Trial and Erasmus-MC Sophia datasets. Segmentations were manually corrected in 3D Slicer. The corrected airway segmentations were used to retrain the 3D-Unet. Airway measurements were automatically obtained and included count, airway length and luminal diameter per generation from the segmentations. Correcting segmentations required 2–4 h per scan. Manually corrected segmentations had more branches (p < 0.001), longer airways (p < 0.001) and smaller luminal diameters (p = 0.004) than initial segmentations. Segmentations from retrained 3D-Unets trended towards more branches and longer airways compared to the initial segmentations. The largest changes were seen in airways from 6th generation onwards. Manual correction results in significantly improved segmentations and is potentially a useful and time-efficient method to improve the AI tool performance on a specific hospital or research dataset.

Manual Correction of Voxel Misclassifications in Mesiotemporal Structures Does Not Alter Brain–Behavioral Results in an Episodic Memory Task

Voxel-based morphometry (VBM) is an established method for assessing grey matter volumes across the brain. The quality of magnetic resonance imaging (MRI) and the chosen data preprocessing steps can affect the outcome of VBM analyses. We recognized a lack of publicly available and commonly used protocols, which indicates that standardized and optimized preprocessing protocols are needed. This paper focuses on the time- and resource-consuming manual correction of misclassifications of grey matter voxels in cortical structures important in Alzheimer’s dementia. A total of 126 individuals, including 63 patients with very early Alzheimer’s disease and 63 cognitively normal participants, received thorough neuropsychological testing and 3-Tesla MRI. Automated preprocessing of T1 MPRAGE images was performed, and misclassifications of grey matter voxels were manually identified and corrected. In a second run, the manual correction step was skipped. Multiple regression analyses using DARTEL in SPM8 were then conducted with the manually corrected and uncorrected sample, respectively. Manual correction of voxel misclassifications did not have a major impact on the correlation between episodic memory performance and structural brain imaging results. We conclude that, although performing all preprocessing steps remains the gold standard, skipping manual correction of voxel misclassifications is permitted when investigating populations on the Alzheimer’s disease spectrum.

Deep neural networks based automated extraction of dugong feeding trails from UAV images in the intertidal seagrass beds

Dugongs (Dugong dugon) are seagrass specialists distributed in shallow coastal waters in tropical and subtropical seas. The area and distribution of the dugongs’ feeding trails, which are unvegetated winding tracks left after feeding, have been used as an indicator of their feeding ground utilization. However, current ground-based measurements of these trails require a large amount of time and effort. Here, we developed effective methods to observe the dugongs’ feeding trails using unmanned aerial vehicle (UAV) images (1) by extracting the dugong feeding trails using deep neural networks. Furthermore, we demonstrated two applications as follows; (2) extraction of the daily new feeding trails with deep neural networks and (3) estimation the direction of the feeding trails. We obtained aerial photographs from the intertidal seagrass bed at Talibong Island, Trang Province, Thailand. The F1 scores, which are a measure of binary classification model’s accuracy taking false positives and false negatives into account, for the method (1) were 89.5% and 87.7% for the images with ground sampling resolutions of 1 cm/pixel and 0.5 cm/pixel, respectively, while the F1 score for the method (2) was 61.9%. The F1 score for the method (1) was high enough to perform scientific studies on the dugong. However, the method (2) should be improved, and there remains a need for manual correction. The mean area of the extracted daily new feeding trails from September 12–27, 2019, was 187.8 m2 per day (n = 9). Total 63.9% of the feeding trails was estimated to have direction within a range of 112.5° and 157.5°. These proposed new methods will reduce the time and efforts required for future feeding trail observations and contribute to future assessments of the dugongs’ seagrass habitat use.

Bootstrapping Dependency Treebank of Urdu Noisy Text

This paper describes how bootstrapping was used to extend the development of the Urdu Noisy Text dependency treebank. To overcome the bottleneck of manually annotating corpus for a new domain of user-generated text, MaltParser, an opensource, data-driven dependency parser, is used to bootstrap the treebank in semi-automatic manner for corpus annotation after being trained on 500 tweet Urdu Noisy Text Dependency Treebank. Total four bootstrapping iterations were performed. At the end of each iteration, 300 Urdu tweets were automatically tagged, and the performance of parser model was evaluated against the development set. 75 automatically tagged tweets were randomly selected out of pre-tagged 300 tweets for manual correction, which were then added in the training set for parser retraining. Finally, at the end of last iteration, parser performance was evaluated against test set. The final supervised bootstrapping model obtains a LA of 72.1%, UAS of 75.7% and LAS of 64.9%, which is a significant improvement over baseline score of 69.8% LA, 74% UAS, and 62.9% LAS

Laser Scanning and Aerial Photography with UAV in Studying the Structure of Forest-Tundra Stands in the Khibiny Mountains

Studying the structure of stands is a key point in assessing the role of trees in carbon deposition. Information on the spatial structure of ground vegetation at the upper treeline is still insufficiently presented in modern studies. High resolution remote sensing can provide important data to understand the properties and dynamics of vegetation in these conditions. We test the applicability of ground-based mobile laser scanning of the terrain and aerial photography for the rapid and high-precision assessment of the characteristics of tree stands in the forest-tundra ecotone. We obtained canopy height models (CHMs) of the forest and supplemented them with aerial photographs of the research area on the southeastern slope of the Khibiny Mountains. Using CHMs we have delineated boundaries of tree crowns. The height and projection area were determined for each tree. The first characteristic obtained by laser scanning was compared to the heights of the same trees estimated by field measurements. This was done for the purposes of verification. The comparison revealed that laser scanning data allow to set heights closest to field measurements in case the heights are determined by the maximum values of brightness of pixels of CHMs with manual correction of values when outliers are detected (R2 = 0.84). Since manual correction of outliers is time-consuming, we proposed a way to automate the measurements by determining tree heights using the sum of the average value of pixel brightness and the standard deviation multiplied by 2.5 (R2 = = 0.79). We compared the area characteristics of the stands obtained by laser scanning and the unmanned aerial vehicle (UAV) photography. Thus, we obtained detailed information on the spatial location and size of 4424 trees in an area of about 10 ha and compared the results of measuring tree characteristics obtained by different methods. It was also found that with increasing height from 290 to 425 m above sea level on the studied slope, the average height of stands decreases gradually from 4.5–5.0 to 1.1–1.6 m with small fluctuations (0.2–0.4 m), while the density of stands changes from 4620–5860 to 145 m2/ha in a non-linear way.

Reducing Manual Operation Time to Obtain a Segmentation Learning Model for Volume Electron Microscopy Using Stepwise Deep Learning With Manual Correction

Three-dimensional (3D) observation of a biological sample using serial-section electron microscopy is widely used. However, organelle segmentation requires a significant amount of manual time. Therefore, several studies have been conducted to improve their efficiency. One such promising method is 3D deep learning (DL), which is highly accurate. However, the creation of training data for 3D DL still requires manual time and effort. In this study, we developed a highly efficient integrated image segmentation tool that includes stepwise DL with manual correction. The tool has four functions: efficient tracers for annotation, model training/inference for organelle segmentation using a lightweight convolutional neural network, efficient proofreading, and model refinement. We applied this tool to increase the training data step by step (stepwise annotation method) to segment the mitochondria in the cells of the cerebral cortex. We found that the stepwise annotation method reduced the manual operation time by one-third compared with that of the fully manual method, where all the training data were created manually. Moreover, we demonstrated that the F1 score, the metric of segmentation accuracy, was 0.9 by training the 3D DL model with these training data. The stepwise annotation method using this tool and the 3D DL model improved the segmentation efficiency for various organelles.

Impact of additional mattresses in emergency CT on the automated patient centering proposed by a 3D camera: a phantom study

To assess the impact of the use of additional mattresses of different thicknesses on radiation dose and image noise based on the patient centering proposed by a 3D camera for CT. An anthropomorphic phantom was placed on mattresses of different thicknesses (from 3.5 to 13.5 cm) on the table of a CT scanner. The automated patient centering proposed by a 3D camera was analysed as a function of mattress thickness and corrected for table height. For this purpose, the impact on image noise in the lung tissues in the chest area and in the soft tissues in the abdomen-pelvis area, modulated mAs (mAsmod) by the tube current modulation system (TCM) and volume CT dose index (CTDIvol) was assessed slice-by-slice along the z-axis after CT scans. With the use of a mattress, the automated centering proposed by the 3D camera resulted in placement of the phantom above the isocentre. This incorrect positioning led to a significant increase in the mAsmod along the z-axis (p < 0.05) and in the CTDIvol. Image noise was significantly higher (p < 0.05) for automated phantom centering than with manual phantom centering. Differences of image noise between acquisitions with mattresses after automatic and manual phantom centering increased with the mattress thicknesses. The use of an additional mattress placed between the patient’s back and the table-top would require correcting the vertical centering proposed by the 3D camera. This manual correction is essential to avoid increased dose delivered to the patient and higher image noise.