More Like this

1. Core handling, transportation and processing for the South Pole ice core (SPICEcore) project

   https://doi.org/10.1017/aog.2020.80  

   Souney, Joseph M. ; Twickler, Mark S. ; Aydin, Murat ; Steig, Eric J. ; Fudge, T. J. ; Street, Leah V. ; Nicewonger, Melinda R. ; Kahle, Emma C. ; Johnson, Jay A. ; Kuhl, Tanner W. ; et al ( January 2020 , Annals of Glaciology)

   null (Ed.)

   Abstract An intermediate-depth (1751 m) ice core was drilled at the South Pole between 2014 and 2016 using the newly designed US Intermediate Depth Drill. The South Pole ice core is the highest-resolution interior East Antarctic ice core record that extends into the glacial period. The methods used at the South Pole to handle and log the drilled ice, the procedures used to safely retrograde the ice back to the National Science Foundation Ice Core Facility (NSF-ICF), and the methods used to process and sample the ice at the NSF-ICF are described. The South Pole ice core exhibited minimal brittle ice, which was likely due to site characteristics and, to a lesser extent, to drill technology and core handling procedures. 

   more » « less
2. The Temple University Digital Pathology Corpus: The Breast Tissue Subset

   https://doi.org/10.1109/SPMB52430.2021.9672275  

   Wevodau, Z. ; Doshna, B. ; Jhala, N. ; Akhtar, I. ; Obeid, I. ; Picone, J. ( December 2021 , Proceedings of the IEEE Signal Processing in Medicine and Biology Symposium (SPMB))

   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 [1], as part of its National Science Foundation-funded Major Research Instrumentation grant titled “MRI: High Performance Digital Pathology Using Big Data and Machine Learning” [2]. 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 [3], image recognition [4] and text processing [5] 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 [6]. A preliminary version of this breast corpus release was tested in a pilot study using a baseline machine learning system, ResNet18 [7], 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 [1] using the nine labels in Table 1 [8] 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 [9]. 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 [1] 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. [2] 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/. [3] 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. [4] 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. [5] 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]. [6] 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/. [7] 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/. [8] 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. [9] 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
3. Theory to Practice: Professional Development for Culturally Responsive Technician Education

   Pickering, C. ( July 2022 , ASEE Annual Conference proceedings)

   The HSI (Hispanic Serving Institution) ATE (Advanced Technological Education) Hub 2 is a three-year collaborative research project funded by the National Science Foundation (NSF) that continues the partnership between two successful programs and involves a third partner in piloting professional development that draws upon findings from the initial program. The goal of HSI ATE Hub 2 is to improve outcomes for Latinx students in technician education programs through design, development, pilot delivery, and dissemination of a 3-tier professional development (PD) model for culturally responsive technician education at 2-year Hispanic Serving Institutions (HSIs). The project seeks to do this by developing the awareness and ability of faculty to appreciate, engage, and affirm the unique cultural identities of the students in their classes and use this connection to deepen students’ belonging and emerging identities as STEM learners and future STEM technicians. This paper shares the research foundations shaping this approach and the methods by which faculty professional development is being provided to develop this important and sensitive instructional capability in participating faculty. The tiered PD model features a scaffolded series of reflective and activity-oriented modules to incrementally enrich the instructional practices and mindset of HSI STEM educators and strengthen their repertoire of strategies for engaging culturally diverse students. Scaffolding that translates culturally responsive theory to practice spans each of the four distinct topic modules in each tier. Each topic module in a tier then scaffolds to a more advanced topic module in the next tier. Tier 1, Bienvenidos, welcomes HSI STEM educators who recognize the need to better serve their Latinx students, and want guidance for small practical activities to try with their students. Tier 2, Transformation through Action, immerses HSI STEM educators in additional activities that bring culturally responsive practices into their technician training while building capacity to collect evidence about impacts and outcomes for students. Tier 3, Engaging Community, strengthens leadership as HSI STEM educators disseminate results from activities completed in Tiers 1 and 2 at conferences that attract technician educators. Sharing the evidence-based practices and their outcomes contributes to achieving broader impacts in the Advanced Technological Education or ATE Community of NSF grantees. Westchester Community College (WCC), the first 2-year HSI in the State University of New York (SUNY) 64 campus system, is piloting the 3-tier PD model using virtual learning methods mastered through previous NSF ATE work and the COVID-19 context. During the pilot, over 20 WCC technician educators in three cohorts will develop leadership skills and practice culturally responsive methods. The pilot will build capacity within WCC STEM technician programs to better support the diversity of students, industry demand for a diverse workforce, and WCC’s capacity for future development of technician education programs. This first paper in a three part series describes the program goals and objectives, the 3-Tier PD model, and reports initial results for Cohort A’s engagement in the first three modules of Tier 1. 

   more » « less
4. Theory to Practice: Professional Development for Culturally Responsive Technician Education

   Pickering, C. ; Miller McNeill, L. ; Lopez, M. ; Rodriguez, J. R. ; Belknap, S. ; Craft, E. L. ; VanIngen-Dunn, C. ( July 2022 , ASEE Annual Conference proceedings)

   The HSI (Hispanic Serving Institution) ATE (Advanced Technological Education) Hub 2 is a three-year collaborative research project funded by the National Science Foundation (NSF) that continues the partnership between two successful programs and involves a third partner in piloting professional development that draws upon findings from the initial program. The goal of HSI ATE Hub 2 is to improve outcomes for Latinx students in technician education programs through design, development, pilot delivery, and dissemination of a 3-tier professional development (PD) model for culturally responsive technician education at 2-year Hispanic Serving Institutions (HSIs). The project seeks to do this by developing the awareness and ability of faculty to appreciate, engage, and affirm the unique cultural identities of the students in their classes and use this connection to deepen students’ belonging and emerging identities as STEM learners and future STEM technicians. This paper shares the research foundations shaping this approach and the methods by which faculty professional development is being provided to develop this important and sensitive instructional capability in participating faculty. The tiered PD model features a scaffolded series of reflective and activity-oriented modules to incrementally enrich the instructional practices and mindset of HSI STEM educators and strengthen their repertoire of strategies for engaging culturally diverse students. Scaffolding that translates culturally responsive theory to practice spans each of the four distinct topic modules in each tier. Each topic module in a tier then scaffolds to a more advanced topic module in the next tier. Tier 1, Bienvenidos, welcomes HSI STEM educators who recognize the need to better serve their Latinx students, and want guidance for small practical activities to try with their students. Tier 2, Transformation through Action, immerses HSI STEM educators in additional activities that bring culturally responsive practices into their technician training while building capacity to collect evidence about impacts and outcomes for students. Tier 3, Engaging Community, strengthens leadership as HSI STEM educators disseminate results from activities completed in Tiers 1 and 2 at conferences that attract technician educators. Sharing the evidence-based practices and their outcomes contributes to achieving broader impacts in the Advanced Technological Education or ATE Community of NSF grantees. Westchester Community College (WCC), the first 2-year HSI in the State University of New York (SUNY) 64 campus system, is piloting the 3-tier PD model using virtual learning methods mastered through previous NSF ATE work and the COVID-19 context. During the pilot, over 20 WCC technician educators in three cohorts will develop leadership skills and practice culturally responsive methods. The pilot will build capacity within WCC STEM technician programs to better support the diversity of students, industry demand for a diverse workforce, and WCC’s capacity for future development of technician education programs. This first paper in a three part series describes the program goals and objectives, the 3-Tier PD model, and reports initial results for Cohort A’s engagement in the first three modules of Tier 1. 

   more » « less
5. Multidisciplinary Research and Teaching by Means of Employing FTIR Spectroscopic Imaging System and Characterization Techniques

   https://doi.org/10.18260/1-2--34988  

   Alavi, Zahrasadat ( June 2020 , 2020 ASEE Virtual Annual Conference Content Access Proceedings)

   null (Ed.)

   This paper focuses on discussing the efforts made to engage students in multi-disciplinary research and integrate teaching and research in the areas of FTIR Spectro- microscopy and image processing and analysis. The author (PI) and co-PIs acquired a Fourier Transform Infrared (FTIR) Spectroscopic Imaging equipment through the National Science Foundation- Major Research Instrumentation (NSF- MRI) grant (#1827134). This project aims to use the equipment to conduct undergraduate and graduate research projects and teach undergraduate and graduate classes. The NSF awarded the California State University Chico (CSU Chico) $175,305 to acquire an FTIR spectrometer and microscope, which are important tools for chemical characterization of samples with infrared active molecules. FTIR Spectroscopic Imaging System especially provides accurate chemical images that reveal the variations in images’ pixels which are mappings of constituent materials of samples rather than a single visible image with slight variations. By employing this equipment in research and the Image Processing course, students can learn how to collect, process and analyze the imaging data of samples and the corresponding spectral data. The students not only will learn how to process a single chemical image, but also will work with the data cubes to consider the pixel intensities along the IR spectrum, experience working with big data, hone the skills to design experiments, analyze larger data sets, develop pre- and post-image processing techniques, and apply and refine math and programming skills. Image processing course conventionally is based on math, digital signal and systems, and requires programming skills such as Matlab, C++, and Python. along with the mentioned knowledge. Additionally, the research conducted by this equipment promotes collaboration between engineering major students and science major students. In this paper, the author will explain how collecting data through running experiments with the FTIR Spectroscopic Imaging equipment helps students visualize theory and relate it to real world problems. This paper also discusses the results of engaging undergraduate students from various majors in research. Moreover, it will discuss some of the projects that were conducted by undergraduate students and their learning outcomes. The objective of the research projects was material characterization towards contribution to health by employing FTIR Spectroscopic Imaging System. 

   more » « less