Semi-automatic Method Research Articles

A Novel Remote Sensing Image Registration Algorithm Based on Feature Using ProbNet-RANSAC.

Image registration based on feature is a commonly used approach due to its robustness in complex geometric deformation and larger gray difference. However, in practical application, due to the effect of various noises, occlusions, shadows, gray differences, and even changes of image contents, the corresponding feature point set may be contaminated, which may degrade the accuracy of the transformation model estimate based on Random Sample Consensus (RANSAC). In this work, we proposed a semi-automated method to create the image registration training data, which greatly reduced the workload of labeling and made it possible to train a deep neural network. In addition, for the model estimation based on RANSAC, we determined the process according to a probabilistic perspective and presented a formulation of RANSAC with the learned guidance of hypothesis sampling. At the same time, a deep convolutional neural network of ProbNet was built to generate a sampling probability of corresponding feature points, which were then used to guide the sampling of a minimum set of RANSAC to acquire a more accurate estimation model. To illustrate the effectiveness and advantages of the proposed method, qualitative and quantitative experiments are conducted. In the qualitative experiment, the effectiveness of the proposed method was illustrated by a checkerboard visualization of image pairs before and after being registered by the proposed method. In the quantitative experiment, other three representative and popular methods of vanilla RANSAC, LMeds-RANSAC, and ProSAC-RANSAC were compared, and seven different measures were introduced to comprehensively evaluate the performance of the proposed method. The quantitative experimental result showed that the proposed method had better performance than the other methods. Furthermore, with the integration of the model estimation of the image registration into the deep-learning framework, it was possible to jointly optimize all the processes of image registration via end-to-end learning to further improve the accuracy of image registration.

A systematic method for identifying references to academic research in grey literature.

Grey literature encompasses documents not published in academic journals or books. Some grey literature has substantial societal importance, such as medical guidelines, government analyses and pressure group reports. Academic research cited in such documents may therefore have had indirect societal impact, such as in policy making, clinical practice or legislation. Identifying citations to academic research from grey literature may therefore help assess its societal impacts. This is difficult, however, due to the variety of document and referencing formats used in grey literature, even from a single organisation. In response, this study introduces and tests a semi-automatic method to match academic journal articles with unstandardised grey literature cited references. For this, the metadata (lead author last name, title, year) of 2.45 million UK Russell Group university outputs was matched against a 100-document sample of UK government grey literature to assess the accuracy of 21 matching heuristics. The optimal method (lead author last name and title in either order, maximum of 200 characters apart) is sufficiently accurate and scalable to make the task of matching research outputs to grey literature references feasible. The method was then applied to 3347 government publications, showing approximately 23% of UK government grey literature in this study contained at least one reference to UK Russell Group university output, and of this grey literature, an average of 3.79 references were present per document. The applied method also shows that economics and environmental science academic research is most cited between 2010 and 2018.Supplementary InformationThe online version contains supplementary material available at 10.1007/s11192-022-04408-4.

Decreased steroidogenic enzyme activity in benign adrenocortical tumors is more pronounced in bilateral lesions as determined by steroid profiling in LC-MS/MS during ACTH stimulation test.

ObjectiveLarge response of steroid precursors, including 17-hydroxyprogesterone, to adrenocorticotropic hormone (ACTH) has been described in adrenocortical tumors, suggesting the existence of intra-tumoral enzymatic deficiencies. This study aimed to compare steroidogenesis enzymes activity in unilateral and bilateral benign tumors using serum steroid profiling in liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) in the basal state and after ACTH 1-24 stimulation.Design and methodsA serum profile of seven consecutive adrenal steroids was determined in LC-MS/MS in the basal state (T0) and after ACTH 1-24 stimulation (T60) in 35 patients with bilateral adrenocortical tumors (BL), 38 patients with unilateral tumors (UL) and 37 control subjects (CT). Response amplitude of each individual steroid was evaluated by T60/T0 ratio, whereas enzymatic activity was assessed by the downstream/upstream steroid ratio. Adrenal volume was quantified by a semi-automatic segmentation method.ResultsFor the seven steroids assayed, the amplitude of response to ACTH was higher in BL than in UL and in CT. The difference between BL and UL persisted even after matching patients on adrenal volume. On glucocorticoids pathway, enzymatic activity of CYP11B1 was significantly decreased in BL (78.3 (43.1-199.4)) in comparison to both UL (122.7 (13.8-228.4), P = 0.0002) and CT (186.8 (42.1-1236.3), P < 0.0001). On mineralocorticoids and androgens pathways, the enzymatic activity of CYP11B2 and CYP17A1-17,20 lyase was also lower in BL than UL and CT.ConclusionsDecreased activity of distal steroidogenesis enzymes CYP11B1, CYP11B2 and CYP17A1-17,20 lyase, responsible for an explosive response to ACTH of upstream precursors in bilateral tumors, limits the synthesis of bioactive steroids, in particular cortisol, despite the increase in adrenal mass.Significance statementActivity of distal steroidogenesis enzymes (CYP11B1, CYP11B2 and CYP17A1 on glucocorticoids, mineralocorticoids and androgens pathways, respectively) is decreased in adrenocortical benign tumors. This decrease is more pronounced in bilateral lesions and seems to depend more on the nature of the lesion than on the increase in adrenal volume. It is responsible for the explosive response to ACTH of steroid precursors located upstream of these enzymes. It probably allows bioactive steroids, particularly cortisol, to stay in the normal range for a long time despite the increase in adrenal mass.

Prediction of Implant Size Based on Breast Volume Using Mammography with Fully Automated Measurements and Breast MRI.

Determination of implant size is crucial for patients with breast cancer undergoing one-stage breast reconstruction. The purpose of this study is to predict the implant size based on the breast volume measured by mammography (MG) with a fully automated method, and by breast magnetic resonance imaging (MRI) with a semi-automated method, in breast cancer patients with direct-to-implant reconstruction. This retrospective study included 84 patients with breast cancer who underwent direct-to-implant reconstruction after nipple-sparing or skin-sparing mastectomy and preoperative MG and MRI between April 2015 and April 2019. Breast volume was measured using (a) MG with a fully automated commercial software and (b) MRI with an in-house semi-automated software program. Multivariable regression analyses including breast volume and patient weight (P < 0.05 in univariable analysis) were conducted to predict implant size. MG and MRI breast volume was highly correlated with both implant size (correlation coefficient 0.862 and 0.867, respectively; P values < 0.001) and specimen weight (correlation coefficient 0.802 and 0.852, respectively; P values < 0.001). Mean absolute difference between the MR breast volume and implant size was 160 cc, which was significantly higher than that between the MG breast volume and implant size of 118 cc (P < 0.001). On multivariable analyses, only breast volume measured by both MG and MRI was significantly associated with implant size in any implant type (all P values < 0.001). Breast volume measured by MG and MRI can be used to predict appropriate implant size in breast cancer patients undergoing direct-to-implant reconstruction in an efficient and objective manner.

High-Throughput Imaging of Blood Flow Reveals Developmental Changes in Distribution Patterns of Hemodynamic Quantities in Developing Zebrafish.

Mechanical forces from blood flow and pressure (hemodynamic forces) contribute to the formation and shaping of the blood vascular network during embryonic development. Previous studies have demonstrated that hemodynamic forces regulate signaling and gene expression in endothelial cells that line the inner surface of vascular tubes, thereby modifying their cellular state and behavior. Given its important role in vascular development, we still know very little about the quantitative aspects of hemodynamics that endothelial cells experience due to the difficulty in measuring forces in vivo. In this study, we sought to determine the magnitude of wall shear stress (WSS) exerted on ECs by blood flow in different vessel types and how it evolves during development. Utilizing the zebrafish as a vertebrate model system, we have established a semi-automated high-throughput fluorescent imaging system to capture the flow of red blood cells in an entire zebrafish between 2- and 6-day post-fertilization (dpf). This system is capable of imaging up to 50 zebrafish at a time. A semi-automated analysis method was developed to calculate WSS in zebrafish trunk vessels. This was achieved by measuring red blood cell flow using particle tracking velocimetry analysis, generating a custom-made script to measure lumen diameter, and measuring local tube hematocrit levels to calculate the effective blood viscosity at each developmental stage. With this methodology, we were able to determine WSS magnitude in different vessels at different stages of embryonic and larvae growth and identified developmental changes in WSS, with absolute levels of peak WSS in all vessel types falling to levels below 0.3 Pa at 6 dpf. Additionally, we discovered that zebrafish display an anterior-to-posterior trend in WSS at each developmental stage.

The AlpArray Research Seismicity-Catalogue

Summary We take advantage of the new large AlpArray Seismic Network (AASN) as part of the AlpArray research initiative (www.alparray.ethz.ch), to establish a consistent seismicity-catalogue for the greater Alpine region (GAR) for the time period 2016 January 1–2019 December 31. We use data from 1103 stations including the AASN backbone composed of 352 permanent and 276 (including 30 OBS) temporary broad-band stations (network code Z3). Although characterized by a moderate seismic hazard, the European Alps and surrounding regions have a higher seismic risk due to the higher concentration of values and people. For these reasons, the GAR seismicity is monitored and routinely reported in catalogues by a 11 national and 2 regional seismic observatories. The heterogeneity of these data set limits the possibility of extracting consistent information by simply merging to investigate the GAR's seismicity as a whole. The uniformly spaced and dense AASN provides, for the first time, a unique opportunity to calculate high-precision hypocentre locations and consistent magnitude estimation with uniformity and equal uncertainty across the GAR. We present a new, multistep, semi-automatic method to process ∼50 TB of seismic signals, combining three different software. We used the SeisComP3 for the initial earthquake detection, a newly developed Python library ADAPT for high-quality re-picking, and the well-established VELEST algorithm both for filtering and final location purposes. Moreover, we computed new local magnitudes based on the final high-precision hypocentre locations and re-evaluation of the amplitude observations. The final catalogue contains 3293 seismic events and is complete down to local magnitude 2.4 and regionally consistent with the magnitude 3+ of national catalogues for the same time period. Despite covering only 4 yr of seismicity, our catalogue evidences the main fault systems and orogens’ front in the region, that are documented as seismically active by the EPOS-EMSC manually revised regional bulletin for the same time period. Additionally, we jointly inverted for a new regional minimum 1-D P-wave velocity model for the GAR and station delays for both permanent station networks and temporary arrays. These results provide the base for a future re-evaluation of the past decades of seismicity, and for the future seismicity, eventually improving seismic-hazard studies in the region. Moreover, we provide a unique, consistent seismic data set fundamental to further investigate this complex and seismically active area. The catalogue, the minimum 1-D P-wave velocity model, and station delays associated are openly shared and distributed with a permanent DOI listed in the data availability section.

Automatic defect detection of texture surface with an efficient texture removal network

Automated defect inspection of texture surface is still a challenging task in the industrial automation field due to the tremendous changes in the appearance of various surface textures. We present a simple but powerful image transformation network to remove textures and highlight defects at full resolution. The simple full convolution network consists only of 3 × 3 regular convolution and several dilated convolution blocks, which makes it compact and able to capture multi-scale features effectively. To further improve the ability of the network to suppress texture and highlight defects, a polynomial loss function combining perceptual loss, structural similarity loss and image gradient loss is proposed. In addition, a semi-automatic annotation method mainly composed of wavelet transform and relative total variation is designed to generate a data set of image pairs containing the original texture image and the desired texture removal image. We conducted experiments on a milled metal surface defect dataset and an open data set containing various textured backgrounds to evaluate the performance of our method. Compared with other convolutional neural network approaches, the results demonstrate the superiority of the proposed method. The method has been applied to the surface defect online detection system of an aluminum ingot milling production line, which effectively improves the surface inspection efficiency and product quality.

Threshold field painting saves the time for segmentation of minute arteries.

It is often time-consuming to segment fine structures, such as the cerebral arteries from magnetic resonance imaging (MRI). Moreover, extracting anatomically abnormal structures is generally difficult. The segmentation workflow called threshold field painting was tested for its feasibility in morbid minute artery segmentation with special emphasis on time efficiency. Seven patients with meningioma with ten-sided feeding arteries (n = 10) originating from middle meningeal arteries (MMA) were investigated by three experts of the conventional method for segmentation. The MRI time-of-flight sequence was utilized for the segmentation of each procedure. The tasks were accomplished using both the conventional method and the proposed method in random order. The task completion time and usability score were analyzed using the Wilcoxon signed-rank test. Except for one examinee (P = 0.06), the completion time significantly decreased (both P < 0.01) with the use of the proposed method. The average task completion time among the three examinees for the conventional method was 2.8 times longer than that for the proposed method. The usability score was generally in favor of the proposed method. The normally nonexistent minute arteries, such as the MMA feeders, were deemed more efficiently segmented with the proposed method than with the conventional method. While automatic segmentation might be the ultimate solution, our semiautomatic method incorporating expert knowledge is expected to work as the practical solution.

Homogeneity assessment of Swiss snow depth series: comparison of break detection capabilities of (semi-)automatic homogenization methods

Abstract. Knowledge concerning possible inhomogeneities in a data set is of key importance for any subsequent climatological analyses. Well-established relative homogenization methods developed for temperature and precipitation exist but have rarely been applied to snow-cover-related time series. We undertook a homogeneity assessment of Swiss monthly snow depth series by running and comparing the results from three well-established semi-automatic break point detection methods (ACMANT – Adapted Caussinus-Mestre Algorithm for Networks of Temperature series, Climatol – Climate Tools, and HOMER – HOMogenizaton softwarE in R). The multi-method approach allowed us to compare the different methods and to establish more robust results using a consensus of at least two change points in close proximity to each other. We investigated 184 series of various lengths between 1930 and 2021 and ranging from 200 to 2500 m a.s.l. and found 45 valid break points in 41 of the 184 series investigated, of which 71 % could be attributed to relocations or observer changes. Metadata are helpful but not sufficient for break point verification as more than 90 % of recorded events (relocation or observer change) did not lead to valid break points. Using a combined approach (two out of three methods) is highly beneficial as it increases the confidence in identified break points in contrast to any single method, with or without metadata.

NODeJ: an ImageJ plugin for 3D segmentation of nuclear objects

BackgroundThe three-dimensional nuclear arrangement of chromatin impacts many cellular processes operating at the DNA level in animal and plant systems. Chromatin organization is a dynamic process that can be affected by biotic and abiotic stresses. Three-dimensional imaging technology allows to follow these dynamic changes, but only a few semi-automated processing methods currently exist for quantitative analysis of the 3D chromatin organization.ResultsWe present an automated method, Nuclear Object DetectionJ (NODeJ), developed as an imageJ plugin. This program segments and analyzes high intensity domains in nuclei from 3D images. NODeJ performs a Laplacian convolution on the mask of a nucleus to enhance the contrast of intra-nuclear objects and allow their detection. We reanalyzed public datasets and determined that NODeJ is able to accurately identify heterochromatin domains from a diverse set of Arabidopsis thaliana nuclei stained with DAPI or Hoechst. NODeJ is also able to detect signals in nuclei from DNA FISH experiments, allowing for the analysis of specific targets of interest.Conclusion and availabilityNODeJ allows for efficient automated analysis of subnuclear structures by avoiding the semi-automated steps, resulting in reduced processing time and analytical bias. NODeJ is written in Java and provided as an ImageJ plugin with a command line option to perform more high-throughput analyses. NODeJ can be downloaded from https://gitlab.com/axpoulet/image2danalysis/-/releases with source code, documentation and further information avaliable at https://gitlab.com/axpoulet/image2danalysis. The images used in this study are publicly available at https://www.brookes.ac.uk/indepth/images/ and https://doi.org/10.15454/1HSOIE.

Performance of a rule-based semi-automated method to optimize chart abstraction for surveillance imaging among patients treated for non-small cell lung cancer

BackgroundWe aim to develop and test performance of a semi-automated method (computerized query combined with manual review) for chart abstraction in the identification and characterization of surveillance radiology imaging for post-treatment non-small cell lung cancer patients.MethodsA gold standard dataset consisting of 3011 radiology reports from 361 lung cancer patients treated at the Veterans Health Administration from 2008 to 2016 was manually created by an abstractor coding image type, image indication, and image findings. Computerized queries using a text search tool were performed to code reports. The primary endpoint of query performance was evaluated by sensitivity, positive predictive value (PPV), and F1 score. The secondary endpoint of efficiency compared semi-automated abstraction time to manual abstraction time using a separate dataset and the Wilcoxon rank-sum test.ResultsQuery for image type demonstrated the highest sensitivity of 85%, PPV 95%, and F1 score 0.90. Query for image indication demonstrated sensitivity 72%, PPV 70%, and F1 score 0.71. The image findings queries ranged from sensitivity 75–85%, PPV 23–25%, and F1 score 0.36–0.37. Semi-automated abstraction with our best performing query (image type) improved abstraction times by 68% per patient compared to manual abstraction alone (from median 21.5 min (interquartile range 16.0) to 6.9 min (interquartile range 9.5), p < 0.005).ConclusionsSemi-automated abstraction using the best performing query of image type improved abstraction efficiency while preserving data accuracy. The computerized query acts as a pre-processing tool for manual abstraction by restricting effort to relevant images. Determining image indication and findings requires the addition of manual review for a semi-automatic abstraction approach in order to ensure data accuracy.

NFB-08. TRAM-01: A Phase 2 study of trametinib for pediatric patients with neurofibromatosis type 1 and plexiform neurofibromas

BACKGROUND: Plexiform neurofibromas (PN) are found in up to 50% of patients with neurofibromatosis type 1 (NF1). Trametinib has been used widely to treat PN but limited data has been reported on its efficacy within a clinical trial. METHODS: This ongoing multicenter phase II trial includes patients with pediatric low-grade glioma and PN. The primary objective for PN was to evaluate the overall response rate based on RECIST 1.1 criteria after daily oral trametinib administration for eighteen 28-day cycles. The volumes of PN were centrally quantified using a new semi-automatic 3D segmentation method. RESULTS: As of January 1, 2022, 45 patients with PN were enrolled in the study. Twenty-eight completed treatment and were available for analysis. For these patients, the median age was 11.4 years (range 0.7-19.8) including 16 males (57.1%). The majority did not receive prior systemic therapies (71.4%). The median volume of PN at baseline was 49.5 cm3 (range 2.6 to 469). Among the 28 patients, 25 (89.3%) completed 18 cycles as planned. One patient discontinued due to adverse reaction, one patient refused to continue treatment and one patient discontinued treatment based on physician decision. Median duration of treatment was 15.9 months (range 4.6 to 16.8). Median duration of follow-up was 29.7 months (range 17.7 to 38.1). A total of 32 PN were available for volumetric analysis. Using RECIST evaluation, the overall response rate was 24.1%. Volumetric assessment demonstrated an overall response rate of 60.7% and 62.5% of PN showed a decrease of more than 20% in volume. Median decrease in volume was -30% (range -93.5 to 14.3). Twenty-seven patients (93.1%) had durable response without progression (lasting ≥1 year). CONCLUSION: We report outcome and volumetric quantification of PN treated with trametinib within a large clinical trial. Based on the current results, trametinib appears effective and offers durable response.

Levering internet-based discussion data from gynecological cancer survivors to inform survivorship care: A mixed methods thematic analysis approach.

e24143 Background: Internet and social media platforms offer insight into the experience of survivorship, but the volume of narrative data is often cumbersome for thorough thematic analysis. We present a semi-automated mixed methods approach to analyze web-based discussions and demonstrate its utility on American Cancer Society (ACS) Discussion Board posts from gynecologic cancer survivors. Methods: Posts from the ACS Ovarian Cancer, Uterine Cancer, and Other Gynecological Cancer Discussion Boards were analyzed for psychosocial themes described in the Cancer Survivorship Care Framework. Quantitative analysis was facilitated by verified keywords created for each theme of interest and was completed via custom Python scripts and R packages. Qualitative analysis was expedited by theme keywords and completed on a subset of posts. Results: A total of 125,498 posts made by 6,436 gynecological cancer survivors and caregivers between July 2000 and February 2020 were evaluated. Of all posts, 23,458 were related to psychosocial aspects of survivorship and were analyzed. Quantitative analysis revealed that survivors frequently discussed the role of friends and family in care, fatigue, health insurance issues, and the effect of cancer on loved ones. The most frequently used psychosocial-related words were “family,” “hope,” and “help.” Qualitative analysis demonstrated that survivors frequently discussed coping and caregiver relationships, where survivors sought advice on topics including spousal emotional adjustment, financial toxicity, and confronting mortality. Conclusions: Online platforms offer great opportunity to learn about patient experiences of survivorship care. Candid conversations between survivors may be leveraged to inform survivorship practices and improve future survivorship initiatives. Our novel methodology expedites thematic analysis of such robust narrative data and its translation into survivorship settings.

Modern challenges of property market analysis- homogeneous areas determination

Effective comprehension of highly complex and spatially heterogenous property market requires its’ appropriate recognition. One of the most critical steps in property analyses and valuation procedures is the identification of the sub-markets as the fundamental comparable units. The biggest challenge in this case is to define the criteria of the basis indicating the similarity (homogeneity) of property markets area. The objective of the study was to propose methodology (called “HO-MAR”) that enables objective identification of the homogenous (similar/comparable) areas (zones) that could indicate location of similar groups of property transactions (representative properties) used either in individual valuations or AVMs. The authors propose utilization of automated solutions based on robust geo-estimation that enables high efficacy of property submarkets identification. Robust geo-estimation is guaranteed by merging semi-automated data mining methods (e.g. entropy theory, rough set theory, fuzzy logic) and geoprocessing activities (Gauss filter, geocoding and reverse geocoding, tessellation model with mutual spatial overlapping) concerning spatial relation database application.

A phenology-based vegetation index classification (PVC) algorithm for coastal salt marshes using Landsat 8 images

• Constructed a new CRVI for red vegetation extraction. • Proposed an enhanced semi-automatic PVC algorithm. • The optimal PVIs reflected the phenological differences of vegetation. • The developed classification model had spatio-temporal stability and scalability. Coastal salt marshes, as a globally significant intertidal ecosystem, are highly productive but extremely fragile and unstable. Mapping coastal salt marshes accurately is the basis of assessing global climate change, biological invasion, and coastal erosion. Using Landsat 8 images, this paper integrated the advantages of pixel- and phenology-based algorithms and vegetation indices in vegetation classification. An enhanced phenology-based vegetation index classification (PVC) algorithm is proposed to obtain the spatial distribution and community composition of coastal salt marshes in Bohai Sea of China accurately and quickly. The results showed that (1) the coastal redness vegetation index (CRVI) can be used to extract Suaeda spp. effectively, and the phenology-based vegetation indices (PVIs) dataset can alleviate the spatial variability of phenology in coastal salt marshes; (2) the crucial phenological periods for identifying coastal salt marshes are May, October, and November, and the optimal PVIs are consistent with the phenological characteristics of salt marshes; (3) during the year 2018–2019, the overall accuracy (OA) of the PVC algorithm in Yancheng coast of Jiangsu Province and Bohai Sea coast reached 80.49 % and 90.8 % respectively. A total of 14,763.39 ha of salt marshes were found in the coastal area of Bohai Sea, and Shandong Province had the most abundant types of salt marshes and the largest area; (4) the classification model based on the PVC algorithm is stable and scalable in 2016–2017 and 2020–2021, with the OA of 89.19% and 86.67% respectively. These results demonstrate the value of the PVC algorithm in vegetation classification, and this study can provide a referable semi-automatic vegetation classification method for other coastal areas.

SELF-TRAINING FOR SEMI-SUPERVISED DEEP CONTOUR DETECTION OF SURFACE WATER

Abstract. Contour detection is better for monitoring dynamic and long-term changes to surface water bodies. For that purpose, we present a semi-automated method for collecting and labeling water contours from Landsat-8 and Sentinel-2 images. Due to the need for human inspection, the method has thus far generated 14K labeled images from more than 1.5M images. Given the cost of data labeling, we propose a deep semi-supervised self-learning system performed in two training stages, known as teacher-student. The teacher is trained on the accurate human-labeled data, then used to pseudo label the remaining unlabeled data. The student is trained on both human-labeled and machine pseudo-labeled data. For both teacher and student, we use a uniquely designed multiscale UNet classifier that uses fewer parameters and is more accurate than other state-of-the-art classifiers. Random augmentations are used to “noise” the student model and improve its generalization, and normalization schemes are used to blend the human-labeled loss with the machine-labeled loss. Comparisons to existing water body detection classifiers and segmentation classifiers show the superiority of our proposed system in detecting water contours.

EVALUATING THE SEPARABILITY BETWEEN DRY TROPICAL FORESTS AND SAVANNA WOODLANDS IN THE BRAZILIAN SAVANNA USING LANDSAT DENSE IMAGE TIME SERIES AND ARTIFICIAL INTELLIGENCE

Abstract. The Brazilian Savanna is the second largest biogeographical region in Brazil and present different vegetation types, consisting mostly of tropical savannas, grasslands, and forests. The forest types have different tree cover and floristic composition, which is associated to leaf deciduousness. Considering the importance of Cerrado to biodiversity conservation and the maintaining of environmental services, the development of methods to map the different forest types in Cerrado is important for conservation programmes, subsidize restauration plains, and to allow estimations of carbon sink and stock. Mapping heterogeneous tropical areas, such as the Brazilian Savanna, is very complex due to the natural factors and peculiarities of the vegetation types, and it's still particularly challenging to separate between different forest formations. In this study we tested machine learning approaches based on the use of dense image time series, in order to evaluate the separability Dry Tropical Forests and Savanna woodlands. We considered the Brazilian State of Tocantins as the study area, which is located in the Northern region of the country. RF classification of Landsat dense time series showed an overall accuracy of 0.85005, while the LSTM approach presented an overall accuracy of 0.88601, with the highest f1-score for the savanna woodlands class, suggesting the capability of the recurrent neural networks on handling complex long-term dependencies such as the EVI dense time series data. This study showed the potential for the development of a semi-automatic method for discriminating the different types of forest formations in the Brazilian Savanna, based on remote sensing.

Detection of Volatile Organic Compounds by Using MEMS Sensors.

We report on the deployment of MEMS static bifurcation (DC) sensors for the detection of volatile organic compounds (VOCs): hydrogen sulfide and formaldehyde. We demonstrate a sensor that can detect as low as a few ppm of hydrogen sulfide. We also demonstrate a sensor array that can selectively detect formaldehyde in the presence of benzene, a closely related interferent. Toward that end, we investigate the sensitivity and selectivity of two detector polymers—polyaniline (PANI) and poly (2,5-dimethyl aniline) (P25DMA)—to both gases. A semiautomatic method is developed to functionalize individual sensors and sensor arrays with the detector polymers. We found that the sensor array can selectively sense 1 ppm of formaldehyde in the presence of benzene.

A convolutional neural network for total tumor segmentation in [64Cu]Cu-DOTATATE PET/CT of patients with neuroendocrine neoplasms

BackgroundSegmentation of neuroendocrine neoplasms (NENs) in [64Cu]Cu-DOTATATE positron emission tomography makes it possible to extract quantitative measures useable for prognostication of patients. However, manual tumor segmentation is cumbersome and time-consuming. Therefore, we aimed to implement and test an artificial intelligence (AI) network for tumor segmentation. Patients with gastroenteropancreatic or lung NEN with [64Cu]Cu-DOTATATE PET/CT performed were included in our training (n = 117) and test cohort (n = 41). Further, 10 patients with no signs of NEN were included as negative controls. Ground truth segmentations were obtained by a standardized semiautomatic method for tumor segmentation by a physician. The nnU-Net framework was used to set up a deep learning U-net architecture. Dice score, sensitivity and precision were used for selection of the final model. AI segmentations were implemented in a clinical imaging viewer where a physician evaluated performance and performed manual adjustments.ResultsCross-validation training was used to generate models and an ensemble model. The ensemble model performed best overall with a lesion-wise dice of 0.850 and pixel-wise dice, precision and sensitivity of 0.801, 0.786 and 0.872, respectively. Performance of the ensemble model was acceptable with some degree of manual adjustment in 35/41 (85%) patients. Final tumor segmentation could be obtained from the AI model with manual adjustments in 5 min versus 17 min for ground truth method, p < 0.01.ConclusionWe implemented and validated an AI model that achieved a high similarity with ground truth segmentation and resulted in faster tumor segmentation. With AI, total tumor segmentation may become feasible in the clinical routine.

Automated Segmentation of Autofluorescence Lesions in Stargardt Disease

To train a deep learning (DL) algorithm to perform fully automated semantic segmentation of multiple autofluorescence lesion types in Stargardt disease. Cross-sectional study with retrospective imaging data. The study included 193 images from 193 eyes of 97 patients with Stargardt disease. Fundus autofluorescence images obtained from patient visits between 2013 and 2020 were annotated with ground-truth labels. Model training and evaluation were performed using fivefold cross-validation. Dice similarity coefficients, intraclass correlation coefficients, and Bland-Altman analyses comparing algorithm-predicted and grader-labeled segmentations. The overall Dice similarity coefficient across all lesion classes was 0.78 (95% confidence interval [CI], 0.69-0.86). Dice coefficients were 0.90 (95% CI, 0.85-0.94) for areas of definitely decreased autofluorescence (DDAF), 0.55 (95% CI, 0.35-0.76) for areas of questionably decreased autofluorescence (QDAF), and 0.88 (95% CI, 0.73-1.00) for areas of abnormal background autofluorescence (ABAF). Intraclass correlation coefficients comparing the ground-truth and automated methods were 0.997 (95% CI, 0.996-0.998) for DDAF, 0.863 (95% CI, 0.823-0.895) for QDAF, and 0.974 (95% CI, 0.966-0.980) for ABAF. A DL algorithm performed accurate segmentation of autofluorescence lesions in Stargardt disease, demonstrating the feasibility of fully automated segmentation as an alternative to manual or semiautomated labeling methods.