- Award ID(s):
- NSF-PAR ID:
- Date Published:
- Journal Name:
- Page Range / eLocation ID:
- 99 to 127
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
Fonseca, Dina (Ed.)Abstract The spread of the Asian tiger mosquito, Aedes albopictus Skuse, throughout the United States has implications for the transmission potential of vector-borne diseases. We used a 30-yr data set of occurrence records in Illinois and developed a hierarchical Bayesian model to shed light on the patterns and processes involved in the introduction and expansion along the northern edge of the geographic range of this species. We also collected specimens from 10 locations and sequenced a segment of their mitochondrial COI genes to assess possible introduction sources and geographic patterns in genetic variation present within contemporary populations. We documented an increase in the number of observations throughout the southern and central parts of Illinois over the study period. The process through which this spread occurred is likely only partially due to local dispersal. The probability of successfully overwintering was likewise low, but both these parameters increased over the study period. This suggests that the presence of Ae. albopictus has been largely due to repeated introductions, but that in recent years populations may have become established and are leading to an increase in locally driven dispersal. There was considerable genetic diversity among populations in Illinois, with 13 distinct haplotypes present in 10 sampling locations, several of which matched haplotypes previously found to be present in locations such as Texas or Japan. Further research is needed to understand how the combination of continued propagule pressure and establishment of populations are driving the increase and expansion of this invasive mosquito along its northern distribution limit.more » « less
Abstract Laboratory and field-based studies of the invasive mosquito Aedes albopictus demonstrate its competency to transmit over twenty different pathogens linked to a broad range of vertebrate hosts. The vectorial capacity of Ae. albopictus to transmit these pathogens remains unclear, partly due to knowledge gaps regarding its feeding behavior. Blood meal analyses from field-captured specimens have shown vastly different feeding patterns, with a wide range of anthropophagy (human feeding) and host diversity. To address this knowledge gap, we asked whether differences in innate host preference may drive observed variation in Ae. albopictus feeding patterns in nature. Low generation colonies (F2–F4) were established with field-collected mosquitoes from three populations with high reported anthropophagy (Thailand, Cameroon, and Florida, USA) and three populations in the United States with low reported anthropophagy (New York, Maryland, and Virginia). The preference of these Ae. albopictus colonies for human versus non-human animal odor was assessed in a dual-port olfactometer along with control Ae. aegypti colonies already known to show divergent behavior in this assay. All Ae. albopictus colonies were less likely (p < 0.05) to choose the human-baited port than the anthropophilic Ae. aegypti control, instead behaving similarly to zoophilic Ae. aegypti . Our results suggest that variation in reported Ae. albopictus feeding patterns are not driven by differences in innate host preference, but may result from differences in host availability. This work is the first to compare Ae. albopictus and Ae. aegypti host preference directly and provides insight into differential vectorial capacity and human feeding risk.more » « less
Abstract A comparative analysis of Raman shifts of quartz inclusions in garnet was made along two traverses across the Connecticut Valley Trough (CVT) in western New England, USA, to examine the regional trends of quartz inclusion in garnet (QuiG) Raman barometry pressure results and to compare this method with conventional thermobarometry and the method of intersecting garnet core isopleths. Overall, Raman shifts of quartz inclusions ranged from 1·2 to 3·5 cm–1 over all field areas and displayed a south to north decrease, matching the overall decrease in mapped metamorphic grade. Raman shifts of quartz inclusions typically did not show systematic variation with respect to their radial position within a garnet crystal, and indicate that garnet probably grew at nearly isothermal and isobaric pressure–temperature (P–T) conditions. The P–T conditions inferred from conventional thermobarometry were in the range of ∼500–575 °C and ∼7·4–10·3 kbar over the sample suite and are in good agreement with previous published thermobarometry throughout the CVT. These P–T results are broadly consistent with QuiG barometry and also suggest that garnet grew isothermally and isobarically at near peak P–T conditions. However, P–T conditions and P–T paths inferred using either garnet core thermobarometry or garnet core intersecting isopleths yield results that are internally inconsistent and generally disagree with the pressure results from QuiG barometry. Garnet core isopleth intersections consistently plotted between the nominal garnet-in curve on mineral assemblage diagrams and the P–T conditions constrained by QuiG isomekes for the majority of the sample suite. Additionally, most samples’ P–T results from QuiG barometry and rim thermobarometry show marked disagreement from those derived from garnet core thermobarometry, compared with the minority that showed agreement within uncertainty. Pressures calculated from QuiG barometry ranged from 8·5 to 9·5 kbar along the traverses in western Massachusetts (MA) and central Vermont (VT) and from 6·5 to 7·5 kbar in northern VT indicating an increase in peak burial of 3–6 km from north to south. Along the western end of the central VT traverse, there are differences in measured Raman shifts and inferred peak pressures of up to 1 kbar across the Richardson Memorial Contact (RMC), indicating a possible fault contact with minor post-peak metamorphic shortening of up to ∼3 km. In contrast, along an east–west traverse in the vicinity of the Goshen Dome, MA, there was little observed variation in Raman shifts across the contact. By contrast, QuiG barometry clearly indicates significant discontinuities in peak pressure east of the Strafford Dome in central VT. This supports the interpretation that post-peak metamorphic shortening was necessary to juxtapose upper staurolite–kyanite zone rocks next to lower garnet zone pelites. Overall, it is concluded that garnet core thermobarometry and garnet core isopleths may provide unreliable results for the P–T conditions of garnet nucleation and inferred P–T paths during garnet growth unless independently verified. The consistency of QuiG results with rim thermobarometry indicates that peak metamorphic conditions previously reported for the CVT using garnet rim thermobarometry are robust and that variation in QuiG barometry results is a valuable tool to analyze structural features within a metamorphic terrane.more » « less
Obeid, I. (Ed.)The Neural Engineering Data Consortium (NEDC) is developing the Temple University Digital Pathology Corpus (TUDP), an open source database of high-resolution images from scanned pathology samples , as part of its National Science Foundation-funded Major Research Instrumentation grant titled “MRI: High Performance Digital Pathology Using Big Data and Machine Learning” . The long-term goal of this project is to release one million images. We have currently scanned over 100,000 images and are in the process of annotating breast tissue data for our first official corpus release, v1.0.0. This release contains 3,505 annotated images of breast tissue including 74 patients with cancerous diagnoses (out of a total of 296 patients). In this poster, we will present an analysis of this corpus and discuss the challenges we have faced in efficiently producing high quality annotations of breast tissue. It is well known that state of the art algorithms in machine learning require vast amounts of data. Fields such as speech recognition , image recognition  and text processing  are able to deliver impressive performance with complex deep learning models because they have developed large corpora to support training of extremely high-dimensional models (e.g., billions of parameters). Other fields that do not have access to such data resources must rely on techniques in which existing models can be adapted to new datasets . A preliminary version of this breast corpus release was tested in a pilot study using a baseline machine learning system, ResNet18 , that leverages several open-source Python tools. The pilot corpus was divided into three sets: train, development, and evaluation. Portions of these slides were manually annotated  using the nine labels in Table 1  to identify five to ten examples of pathological features on each slide. Not every pathological feature is annotated, meaning excluded areas can include focuses particular to these labels that are not used for training. A summary of the number of patches within each label is given in Table 2. To maintain a balanced training set, 1,000 patches of each label were used to train the machine learning model. Throughout all sets, only annotated patches were involved in model development. The performance of this model in identifying all the patches in the evaluation set can be seen in the confusion matrix of classification accuracy in Table 3. The highest performing labels were background, 97% correct identification, and artifact, 76% correct identification. A correlation exists between labels with more than 6,000 development patches and accurate performance on the evaluation set. Additionally, these results indicated a need to further refine the annotation of invasive ductal carcinoma (“indc”), inflammation (“infl”), nonneoplastic features (“nneo”), normal (“norm”) and suspicious (“susp”). This pilot experiment motivated changes to the corpus that will be discussed in detail in this poster presentation. To increase the accuracy of the machine learning model, we modified how we addressed underperforming labels. One common source of error arose with how non-background labels were converted into patches. Large areas of background within other labels were isolated within a patch resulting in connective tissue misrepresenting a non-background label. In response, the annotation overlay margins were revised to exclude benign connective tissue in non-background labels. Corresponding patient reports and supporting immunohistochemical stains further guided annotation reviews. The microscopic diagnoses given by the primary pathologist in these reports detail the pathological findings within each tissue site, but not within each specific slide. The microscopic diagnoses informed revisions specifically targeting annotated regions classified as cancerous, ensuring that the labels “indc” and “dcis” were used only in situations where a micropathologist diagnosed it as such. Further differentiation of cancerous and precancerous labels, as well as the location of their focus on a slide, could be accomplished with supplemental immunohistochemically (IHC) stained slides. When distinguishing whether a focus is a nonneoplastic feature versus a cancerous growth, pathologists employ antigen targeting stains to the tissue in question to confirm the diagnosis. For example, a nonneoplastic feature of usual ductal hyperplasia will display diffuse staining for cytokeratin 5 (CK5) and no diffuse staining for estrogen receptor (ER), while a cancerous growth of ductal carcinoma in situ will have negative or focally positive staining for CK5 and diffuse staining for ER . Many tissue samples contain cancerous and non-cancerous features with morphological overlaps that cause variability between annotators. The informative fields IHC slides provide could play an integral role in machine model pathology diagnostics. Following the revisions made on all the annotations, a second experiment was run using ResNet18. Compared to the pilot study, an increase of model prediction accuracy was seen for the labels indc, infl, nneo, norm, and null. This increase is correlated with an increase in annotated area and annotation accuracy. Model performance in identifying the suspicious label decreased by 25% due to the decrease of 57% in the total annotated area described by this label. A summary of the model performance is given in Table 4, which shows the new prediction accuracy and the absolute change in error rate compared to Table 3. The breast tissue subset we are developing includes 3,505 annotated breast pathology slides from 296 patients. The average size of a scanned SVS file is 363 MB. The annotations are stored in an XML format. A CSV version of the annotation file is also available which provides a flat, or simple, annotation that is easy for machine learning researchers to access and interface to their systems. Each patient is identified by an anonymized medical reference number. Within each patient’s directory, one or more sessions are identified, also anonymized to the first of the month in which the sample was taken. These sessions are broken into groupings of tissue taken on that date (in this case, breast tissue). A deidentified patient report stored as a flat text file is also available. Within these slides there are a total of 16,971 total annotated regions with an average of 4.84 annotations per slide. Among those annotations, 8,035 are non-cancerous (normal, background, null, and artifact,) 6,222 are carcinogenic signs (inflammation, nonneoplastic and suspicious,) and 2,714 are cancerous labels (ductal carcinoma in situ and invasive ductal carcinoma in situ.) The individual patients are split up into three sets: train, development, and evaluation. Of the 74 cancerous patients, 20 were allotted for both the development and evaluation sets, while the remain 34 were allotted for train. The remaining 222 patients were split up to preserve the overall distribution of labels within the corpus. This was done in hope of creating control sets for comparable studies. Overall, the development and evaluation sets each have 80 patients, while the training set has 136 patients. In a related component of this project, slides from the Fox Chase Cancer Center (FCCC) Biosample Repository (https://www.foxchase.org/research/facilities/genetic-research-facilities/biosample-repository -facility) are being digitized in addition to slides provided by Temple University Hospital. This data includes 18 different types of tissue including approximately 38.5% urinary tissue and 16.5% gynecological tissue. These slides and the metadata provided with them are already anonymized and include diagnoses in a spreadsheet with sample and patient ID. We plan to release over 13,000 unannotated slides from the FCCC Corpus simultaneously with v1.0.0 of TUDP. Details of this release will also be discussed in this poster. Few digitally annotated databases of pathology samples like TUDP exist due to the extensive data collection and processing required. The breast corpus subset should be released by November 2021. By December 2021 we should also release the unannotated FCCC data. We are currently annotating urinary tract data as well. We expect to release about 5,600 processed TUH slides in this subset. We have an additional 53,000 unprocessed TUH slides digitized. Corpora of this size will stimulate the development of a new generation of deep learning technology. In clinical settings where resources are limited, an assistive diagnoses model could support pathologists’ workload and even help prioritize suspected cancerous cases. ACKNOWLEDGMENTS This material is supported by the National Science Foundation under grants nos. CNS-1726188 and 1925494. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation. REFERENCES  N. Shawki et al., “The Temple University Digital Pathology Corpus,” in Signal Processing in Medicine and Biology: Emerging Trends in Research and Applications, 1st ed., I. Obeid, I. Selesnick, and J. Picone, Eds. New York City, New York, USA: Springer, 2020, pp. 67 104. https://www.springer.com/gp/book/9783030368432.  J. Picone, T. Farkas, I. Obeid, and Y. Persidsky, “MRI: High Performance Digital Pathology Using Big Data and Machine Learning.” Major Research Instrumentation (MRI), Division of Computer and Network Systems, Award No. 1726188, January 1, 2018 – December 31, 2021. https://www. isip.piconepress.com/projects/nsf_dpath/.  A. Gulati et al., “Conformer: Convolution-augmented Transformer for Speech Recognition,” in Proceedings of the Annual Conference of the International Speech Communication Association (INTERSPEECH), 2020, pp. 5036-5040. https://doi.org/10.21437/interspeech.2020-3015.  C.-J. Wu et al., “Machine Learning at Facebook: Understanding Inference at the Edge,” in Proceedings of the IEEE International Symposium on High Performance Computer Architecture (HPCA), 2019, pp. 331–344. https://ieeexplore.ieee.org/document/8675201.  I. Caswell and B. Liang, “Recent Advances in Google Translate,” Google AI Blog: The latest from Google Research, 2020. [Online]. Available: https://ai.googleblog.com/2020/06/recent-advances-in-google-translate.html. [Accessed: 01-Aug-2021].  V. Khalkhali, N. Shawki, V. Shah, M. Golmohammadi, I. Obeid, and J. Picone, “Low Latency Real-Time Seizure Detection Using Transfer Deep Learning,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB), 2021, pp. 1 7. https://www.isip. piconepress.com/publications/conference_proceedings/2021/ieee_spmb/eeg_transfer_learning/.  J. Picone, T. Farkas, I. Obeid, and Y. Persidsky, “MRI: High Performance Digital Pathology Using Big Data and Machine Learning,” Philadelphia, Pennsylvania, USA, 2020. https://www.isip.piconepress.com/publications/reports/2020/nsf/mri_dpath/.  I. Hunt, S. Husain, J. Simons, I. Obeid, and J. Picone, “Recent Advances in the Temple University Digital Pathology Corpus,” in Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB), 2019, pp. 1–4. https://ieeexplore.ieee.org/document/9037859.  A. P. Martinez, C. Cohen, K. Z. Hanley, and X. (Bill) Li, “Estrogen Receptor and Cytokeratin 5 Are Reliable Markers to Separate Usual Ductal Hyperplasia From Atypical Ductal Hyperplasia and Low-Grade Ductal Carcinoma In Situ,” Arch. Pathol. Lab. Med., vol. 140, no. 7, pp. 686–689, Apr. 2016. https://doi.org/10.5858/arpa.2015-0238-OA.more » « less
On 11 March 2011, the Great East Japan Earthquake triggered a massive tsunami that resulted in the largest known rafting event in recorded history. By spring 2012, marine debris began washing ashore along the Pacific coast of the United States and Canada with a wide range of Asian coastal species attached. We used this unique dataset, where the source region, date of dislodgment and landing location are known, to assess the potential for species invasions by transoceanic rafting on marine debris.
Northeast Pacific from 20 to 60°N.
Major taxa studied
Forty‐eight invertebrate and algal species recorded on Japanese tsunami marine debris (JTMD).
We developed maximum entropy (
MaxEnt) species distribution models for 48 species recorded on JTMD to predict establishment potential along the Pacific coast from 20 to 60°N. Models were compared within the context of historical marine introductions from Japan to this region to validate the emergence of marine debris as a novel vector for species transfer. Results
Overall, 27% (13 species) landed with debris at locations with suitable environmental conditions for establishment and survival, indicating that these species may be able to establish new populations or introduce greater genetic diversity to already established non‐native populations. A further 21 species have an environmental match to areas where tsunami debris likely landed, but was not extensively sampled. Nearly 100 Japanese marine species previously invaded the northeastern Pacific, demonstrating this region’s environmental suitability for rafting Japanese biota. Historical invasions from Japan are highest in California and largely known from bays and harbours.
Marine debris is a novel and growing vector for non‐native species introduction. By utilizing a unique dataset of JTMD species, our predictive models show capacity for new transoceanic invasions and can focus monitoring priorities to detect successful long‐distance dispersal across the world’s oceans.