More Like this

1. Introducing Innovation to First-Year STEM Students through an Intercession Course

   https://doi.org/10.18260/1-2-1153-38332  

   Schubert, Karl D. ; Massey, Leslie B. ; Gattis, Carol S. ( November 2021 , 2021 ASEE Midwest Section Conference Proceedings)

   Innovation training is considered critical for the future of our country, yet despite the important role, opportunities for students to develop innovation skills are limited. For STEM students, training in innovation principles and processes are frequently extra curricular pursuits, such as unpaid internships with start up organizations, shadowing innovation professionals, or obtaining an additional business degree or minor covering innovation principles. The National Science Foundation has funded the authors with a Science, Technology, Engineering and Mathematics (S STEM) grant to provide scholarships combined with research on best practices for recruitment, retention, and development of innovation skills for a diverse group of low income undergraduate students. Students in the program come from STEM disciplines in engineering and the physical sciences however, business students are also integrated into innovation courses although they are not funded by the S STEM grant Design, development, and implementation of the grant funded program’s first innovation related course, a 2 week fall intercession course will be presented Th is first year course is designed to provide the students with an introduction to innovation, develop and nurture the students’ innovation mindset and skills, and also help the students’ successful transition to college. The first-year two-week intercession course was designed and developed with two credit hours focusing on content related to innovation and one credit hour focusing on student success topics. The significant academic course components included: 1) interactive active-learning modules related to innovation processes, identifying where good ideas come from, working in teams, leadership, project management, and communication and presentation skills; 2) team innovation projects, one topic-assigned, applying skills learned in the content modules to develop innovation and team collaboration skills; and 3) integration of business students with STEM students which together gives viewpoints and experiences on product and customer needs. It is important to our nation’s health and safety to instill innovation in our students. In addition, today’s students are interested in innovation and in learning how to apply innovation techniques in their professional and personal lives. The course was designed for teams of four STEM students to one business student which provides a balanced input needed for this type of project taking into account the skillset of the technically oriented STEM students and the marketing-oriented business students, as well as personality types. This ensures that all voices are heard, and topical areas are addressed. There was no problem in getting faculty interest in developing the course, and the collaboration between retention professionals and faculty went well. After the course, an iterative improvement retrospective will be performed on the program as implemented to this point to inform improvements for next year’s cohort. This material is based upon work supported by the National Science Foundation under Grant No. 2030297. 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. 

   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. Use of Personas in Exploring Scholarship Applicants

   Rynearson, Anastasia ; Gartner, Jacqueline ; Miller, Michele ( July 2021 , ASEE Annual Conference proceedings)

   null (Ed.)

   A recent S-STEM award has allowed the engineering program in a rural, liberal arts institution to offer a need-based scholarship program for its students. The engineering program has a number of veteran, underrepresented minority, transfer, and nontraditional students. Many students are also first-generation college students. The institution and engineering program matriculate a number of under-served populations, students who may have needs that are not well understood in the typical engineering education literature. The scholarship program and its associated mentoring and activities will assist workforce development and will also incorporate a number of research avenues to better understand and serve the student population in this unique setting. To apply for the program, students must fill out an application with four 250 – 500 word essay responses relating to their academic progress, perceived barriers to degree completion, and how this award would help them to complete their degree. This study continues work using personas, a method used in human-centered design. Using the first round of scholarship application essays as a source, three personas were developed, one successful applicant, one unsuccessful applicant, and one general applicant. Personas are detailed profiles of a fake person who could reasonably be in each category of interest. In human-centered design, personas are detailed descriptions of likely clients or end-users, developed to help the engineers focus on who they might be designing for. The profiles developed in this study were used to gain insight into which students were likely to choose to apply and which students may be missing out on this opportunity. It is time for another round of applications for this grant and the use of personas will continue and expand as part of this study. Before reviewing applications, the committee will create two personas as ideal candidates instead of developing a standardized rubric. Subsequently, three personas will be developed from the Fall 2020 applications, one for all applicants, one for successful applicants, and one for unsuccessful applicants. These personas will then be compared to the personas created by the application review committee and the personas created from the Fall 2018 applicants. Similarities and differences across the persona groups will be explored to determine whether the applicants are what the reviewers expected and whether the pool of applicants has evolved or remained mostly the same throughout the scholarship program. Review committee members will also be interviewed in a focus group setting to discuss their experiences using common personas rather than standardized rubrics in the application review process. At this time, the applications are not yet due and the analysis has not yet begun. Initial interest for the grant has been strong and we anticipate at least thirty applications for the nineteen available grants. Results presented will include the student profiles and faculty experiences with the use of personas as a metric for reviewing student applications. 

   more » « less
4. Groundwater, soil moisture, light and weather data in Brownsville forest, Nassawadox, VA, 2019-2022

   https://doi.org/10.6073/pasta/942a5a981e6e986c5fa1a9a9cd2eb8b7  

   Fagherazzi, Sergio ; Nordio, Giovanna ( January 2022 , Environmental Data Initiative)

   This dataset includes groundwater, soil moisture, weather and light data collected in six study sites in the Brownsville forested area (VA). In particular, sites H5 and H7 characterize the high forest where healthy Pinus taeda dominate, sites L1 and L6 characterize the low forest, where barren or dead Pinus taeda are present and sites M1 and M2 characterize the medium forest, representing transition between high and low forest. Data collection, started in January 2019, is done for VCR-LTER and CCZN (Coastal Critical Zone Network) long term projects. CTD-Diver are used to measure groundwater pressure, specific conductance and temperature. They hang from a cable in six wells, one for each study site. Water pressure is compensated using barometric pressure data collected by the weather station nearby. Water levels are georeferenced to NAVD 88. One soil moisture sensor for each site is placed 7-10 cm below the ground surface to detect water content, specific conductance and temperature of the first soil layer. A weather station is installed in M1 to collect solar, wind, precipitation and atmospheric pressure data. This material is based upon work supported by the National Science Foundation under Grant No. 2012322, Collaborative Research: Network Cluster: The Coastal Critical Zone: Processes that transform landscapes and fluxes between land and sea. 

   more » « less
5. Finding Resilience in Unexpected Places: Why Design Still Matters in Shrinking Rural Communities

   https://doi.org/10.35483/ACSA.AIA.Inter.21.6  

   Luz, Ana ; Das, Biswa ; Peters, David J. ; Drinkwater, Jennifer ; Zarecor, Kimberly E. ( November 2023 , ACSA Press)

   Quirindongo, Rico ; Theodore, Georgeen (Ed.)

   Most small and rural communities in the United States are shrinking. This population loss is often accompanied by economic and social upheaval—job losses, out migration of young people, school closures, reductions in local services, and deteriorating physical infrastructure. Because design firms cluster in metropolitan areas and most rural commissions are for private clients, architects are largely absent from these places. The AIA Framework for Design Excellence calls for the professional community to enable more sustainable, resilient, and inclusive environments, yet rural places pose a challenge because they remain a strikingly underserved market for architectural services. How can this vision for Design Excellence extend its reach into places where new construction is rare, and architects are not present to learn from and develop relationships with potential clients?This paper presents an overview of an interdisciplinary research project at Iowa State University funded by a $1.5 million grant from the National Science Foundation. The research begins with this question: why do people in some rural towns perceive their quality of life to be increasing even when the population continues to shrink? Using twenty years of survey data about quality of life, the team identified small rural communities in Iowa where the typical association of population loss with community decline did not appear to hold true. Through interviews, site visits, spatial analysis, and data analysis using machine learning and other methods, the team is working to better understand what influences people’s perceptions of quality of life. Understanding more about these unexpectedly resilient communities requires conversations and building trust in places where few outsiders ever visit. Examples of projects in towns working with the research team include adaptive reuse of closed schools and other abandoned properties; improved recreational spaces and parks; and repurposing underused commercial properties.

    

   more » « less