Learning Semantic Image-Text Embeddings in the Radiology Context

Sonit Singh1, Kevin Ho-Shon2, Sarvnaz Karimi3, Len Hamey4

1 Department of Computing, Balaclava Road, NSW, 2109, sonit.singh@hdr.mq.edu.au

2 Macquarie University Hospital, 3 Technology Place, Macquarie University, NSW, 2109, kevin.ho-shon@mq.edu.au

3 Data61, CSIRO, Corner Vimiera & Pembroke Roads, Marsfield NSW, 2122, sarvnaz.karimi@data61.csiro.au

4 Department of Computing, Balaclava Road, NSW, 2109, len.hamey@mq.edu.au

 

Radiologists routinely interpret medical images and describe findings in the form of radiology reports. Inspired by the fact that radiologists often follow templates for writing reports and modify them according to each individual case, we propose a cross-modal retrieval model that aligns visual data (medical images) and textual data (radiology reports) in a shared representation space, allowing retrieval of relevant items that are of different nature with respect to the query format. The model architecture consists of deep neural network where mage features are extracted using off-the-shelf Convolutional Neural Network pre-trained on ImageNet and text features are extracted using multi-scale sentence vectorization methods such as Bag-of-Words (BoW) and Word2Vec. To check the effectiveness of the proposed model, two datasets in the radiology domain, namely, University Chest X-ray collection (IU-CXR) and Radiology Objects in COntext (ROCO) are used. Both datasets consist of medical images and their corresponding captions allowing retrieval of medical image given a text query and vice-versa. For performance evaluation, we report rank-based performance metric Recall@k (where k = 1, 5, 10, 50) which computes the percentage of test images/reports for which at least one correct result is found among the top-k retrieved reports/images. The proposed model not only make radiologists more efficient by retrieving similar cases existing in the Picture Archiving and Communication Systems (PACS), but also allow multi-modal query composition to retrieve medical images or radiology reports as per their specific need.

Mobilising Knowledge in Healthcare Technology Projects through Storytelling

Klaus Veil1 Chivonne Algeo2

1 Western Sydney University, Narellan Rd & Gilchrist Dr, Campbelltown, NSW, 2560, K.Veil@westernsydney.edu.au 

2 Monash University, 900 Dandenong Road, Caulfield East, VIC, 3145, chivonne.algeo@monash.edu

The management of knowledge in technology-driven contexts often neglects the human factors (To Err is Human, NIM, 2000) to enable the delivery of appropriate, and often life-saving outcomes in healthcare settings.  Rather than relying on yet another technical solution, we believe that the active management and dissemination, or ‘mobilisation’, of clinicians’ knowledge during critical junctures through verbal means in healthcare service provision is an essential complement to technology-based systems.

Storytelling can be an effective vehicle to ‘mobilise knowledge’, that is share knowledge and lessons learnt with the goal of achieving successful outcomes.  Telling stories to share knowledge ‘… is just as effective today as any time in history.  People think in terms of metaphors and learn through stories’ (Martin 2000, p. 10).  The capacity to extend an organisation’s capability to make informed, rational decisions can be enhanced by mobilising an individual’s personal knowledge through dialogue, discourse, sharing, and storytelling’ (Dalkir, 2005).  This approach can be applied to the management of technology projects in healthcare, with the aim of improving outcomes and expanding knowledge.

The mobilization of ‘… essential knowledge, including technical knowledge, is often transferred between people by stories, gossip, and by watching one another work.  This is a process in which social interaction is often crucial’ (Pfeffer & Sutton 1999, p. 90). In these interactions, tacit knowledge is often mobilized through conversion.  According to Nonaka, Toyama and Konno (2000, p. 9) explicit knowledge is converted to tacit knowledge, resulting in an anticipated expansion in the quantity and quality of knowledge, in this case for healthcare technology projects.

Initial research has found that developing and sharing research-based products; emphasising brokering; and focusing on implementation are knowledge mobilisation approaches that have been used by healthcare agencies (Davies, Powell & Nutley, 2015).  While participants found a formal evaluation of knowledge mobilisation activities as highly challenging, they reported rich formative experiences.

A recent critical literature review focusing on knowledge management and knowledge mobilisation within healthcare organisations found that while the industry has absorbed some generic concepts, notably Communities of Practice, other more management and performance-oriented perspectives are still developing (Ferlie, Crilly, Jashapara & Peckham, 2012).

Notwithstanding the early stages of adoption of knowledge mobilisation through storytelling and similar approaches, we believe that these resource-efficient methods would benefit the delivery of projects in the healthcare technology domains through an increase in ‘sensemaking’. This approach is how the majority of healthcare workers apply their experience to formally learnt knowledge.  Storytelling offers one way to make sense of what has happened, and may even capture a level of meaning that was only partially grasped before (Mattingly 1991).  This process of sensemaking also involves the ongoing rationalisation of what people are doing (Weick & Sutcliffe 2005).

We will outline how the process of rationalising and organising knowledge enables participants in healthcare technology projects to retrospectively make plausible sense and bring order into ongoing developments, and then use storytelling to mobilise (e.g. pass on) this knowledge to others.  We believe this is a vital activity where healthcare technology and domain-specific knowledge continually and rapidly changes and develops in complex environments.

Learning Analytics in Computer Education Design

Meena Jha1, Sanjay Jha2

1 Central Queensland University, 400 Kent Street, Sydney, NSW 2000, m.jha@cqu.edu.au 

2 Central Queensland University, 400 Kent Street, Sydney, NSW 2000, s.jha@cqu.edu.au

Learning analytics is an emerging field in which sophisticated analytic tools are used to improve learning and education. It draws from, and is closely tied to, a series of other fields of study including business intelligence, web analytics, academic analytics, educational data mining, and action analytics. The field of learning analytics has great potential to inform and enhance teaching and learning practices in higher education. The academic community is seeing current educational systems critically and a lot of issues are being discussed and analyzed. Science, technology, engineering, and mathematics (STEM) fields have notoriously low persistence rates. According to 2016 report on Australia’s STEM workforce only 32% are university qualified. Indeed student retention in STEM discipline is a growing problem. The number of students receiving undergraduate STEM degrees will need to increase as ICT jobs in Australia are expected to grow at a rate of 2% a year upto 2022, more than a third faster that the rate of general job growth. One way to address this problem is by leveraging the emerging field of learning analytics, a data-driven approach to designing learning interventions based on continuously-updated data on learning processes and outcomes.

Learning Management System (LMS) collects user data. It collects all log details. Students and teachers interact with each other via online forums, threaded discussions, and videoconferencing, as well as emails and chats services provided by LMS. These data sets can be analysed to provide answers to: What are the most used resources in computing courses? Who are the most active user in computing courses?. Through an iterative, user-centered, design approach, a learning dashboard can be designed specifically for computing courses. The dashboard can address the important questions relating to learning interventions are: (1) When should the intervention be performed? (2) Whom should the intervention be directed at? (3) What is an effective instructional intervention? LMS provide learners with information content and educational resources. It is an effective way for educators to create, deliver, and manage the educational resources, as well as monitor participation and assess performance among learners.  Learning Analytics (LA), through analysing data and extracting information from LMS can help to find answers to questions that are important to decision-making processes for intervention.  The information on student’s behaviour captured by LMS has been very rarely interrogated and adopted beyond basic load and tool usage. As an educator, we can compare patterns of resource usage, action, or learner’s behaviour by looking at interaction data statements over the term of enrolled user in computing courses. Some aspects of learning such as content engagement, learning preferences, and collaboration can be measured by analysing LMS data. The quantity and diversity of data available regarding student’s online learning behaviour, interactions with peers and teaching staff, and access to other institutional ICT systems (Student services, library, Studiosity, etc..) for example, affords an opportunity for integrating automated student learning and intervention support services. All of these data can be analysed alongside data on students’ learning outcomes in order to identify correlations between learning processes and outcomes, and ultimately to better tailor instruction and intervention to students’ needs. According to our study, students appreciate interventions. Interventions encourage them to get started with the course material, help them prepare for assignments and increase their learning experience.

As an educator need to address (1) How to enable efficient and effective use of teaching resources in computing education; (2) How to identify the need to change and redesign of the computing course assessments regime to ensure all learning objectives are met.

 

 

Convolutional Neural Networks for Prostate Magnetic Resonance Image Segmentation

Tahereh Hassan Zadeh Koohi1, Len Hamey2, Kevin Ho-Shon3

1 Department of Computing, Macquarie University, Balaclava Road, North Ryde, Sydney, NSW, 2109, Australia, tahereh.hassan-zadeh-koohi@hdr.mq.edu.au 

2 Department of Computing, Macquarie University, Balaclava Road, North Ryde, Sydney, NSW, 2109, Australia, len.hamey@mq.edu.au 

3 Department of Computing, Macquarie University, Balaclava Road, North Ryde, Sydney, NSW, 2109, Australia, kevin.ho-shon@mq.edu.au

Digital medical image segmentation is the process of partitioning an image into several discrete and homogeneous regions. Segmentation is needed to find the boundary of the prostate either automatically or semi-automatically. One of the most accurate and non-invasive prostate imaging methods is Magnetic Resonance Imaging (MRI) which is usually employed for the prostate image segmentation and/or possible prostate anomalies detection.

In this research, to improve the Fully Convolutional Neural Network (FCNN) performance for prostate MRI segmentation, we analyse various structures of shortcut connections as well as the size of a deep network. We suggest eight different deep 2D network structures for automatic MRI prostate segmentation based on FCNN.  Our evaluations on the PROMISE12 dataset with ten-fold cross-validation indicate improved and competitive results. We analyse the results in detail, considering MRI slices, MRI volumes, test folds, and also the impact on prostate segmentation of using an EndoRectal Coil to capture the prostate MRI. Our best 2D network outperforms the state-of-the-art 3D FCNN-based methods for prostate MRI segmentation, without any further post-processing module nor pretraining on publicly available data.

 

 

Identifying Microgrids in Rural Areas Using a Genetic Algorithm

Manou Rosenberg1, James Fletcher2, Mark Reynolds3, Lyndon While4, Tim French5

1 The University of Western Australia, 35 Stirling Highway, WA, 6009, manou.rosenberg@research.uwa.edu.au 

2 The University of Western Australia, 35 Stirling Highway, WA, 6009, james.fletcher@research.uwa.edu.au 

3 The University of Western Australia, 35 Stirling Highway, WA, 6009, mark.reynolds@uwa.edu.au 

4 The University of Western Australia, 35 Stirling Highway, WA, 6009, lyndon.while@.uwa.edu.au 

5 The University of Western Australia, 35 Stirling Highway, WA, 6009, tim.french@.uwa.edu.au

The partitioning of networks in order to optimise one or more given objectives is a highly researched topic with many real-world applications, such as dividing water supply networks into so-called district meter areas, or partitioning transport system networks for distributed traffic management. In this research project the aim is to find an optimal distributed network topology for rural electricity networks.

In larger towns and cities the electricity network often has a meshed infrastructure such that most power lines are backed up by other connections in the network. In rural regions an electricity network often spreads over a large area servicing a relatively small number of electricity customers as is the case in many parts of Western Australia. These networks are expensive to build and maintain, and prone to several environmental risks such as bushfires or storms. In the South-West of WA the customers in remote areas are connected to the main grid by long power lines causing a large amount of the infrastructural costs for electricity network operators. As previous research has shown, it can be beneficial to take some of the customers off the interconnected system in order to form nano- or microgrids, where energy is provided by renewable energy sources or small scale generators. The aim is to identify those electricity customers and apply the developed methods to problem instances of Australian electricity customer loads and locations. This could provide a more reliable, cost effective electricity network infrastructure that incorporates a larger percentage of renewable energy resources.

A problem-specific genetic algorithm approach has been developed to determine those network parts, where the electricity customers could be clustered into self-sustaining microgrids. Given a set of electricity customer loads and locations, the microgrid and stand-alone power system formations are identified in order to minimise the total network costs over a certain time period.

The proposed poster will present an overview of background information on electricity networks and microgrids, a brief introduction to microgrids, a description of the evaluated cost function, and an outline of the problem-specific genetic algorithm used for optimising the network. First results for a real-world test instance will be shown as well.

Student Collaboration in IT and Engineering Education

Timothy Boye1

1 University of Technology, Sydney; Faculty of Engineering and IT, Building 11, Floor 5, Office 204, University of Technology, Sydney, P.O. Box 123, Broadway NSW 2007; timothy.d.boye@student.uts.edu.au

Universities are locations of knowledge gathering and creation. Within teaching approaches, collaborative learning is a practice whereby students work together through participation and interaction to synthesise knowledge together (Paulus 2005). Whilst group work is quite popular in technical fields true collaborative learning, as opposed to cooperation, has traditionally been considered easier to implement in fields like the arts rather than in the technical fields due in part to a greater focus on group synthesis tasks in the former and application tasks in the latter. Further, the collaborative tools used in online education environments are touted as the cure-all for implementing collaborative learning, however, collaboration is often not experienced to the fullest extent in these environments and does not happen automatically (e.g. Hathorn & Ingram 2002; Kim 2013).

Previous studies in engineering have shown positive relationships between students’ reporting of their own informal collaboration with their confidence in their learning of course material, knowledge building behaviours, and their course grade (Stump et al 2013). Also gender differences have been found in use of collaborative learning activities (Stump et al 2013) which suggests its use may benefit some underrepresented students. However to do this, learning design, scaffolding and assessment frameworks amongst other factors must be considered by educators for effective collaboration (e.g. Kim 2013; Kurnaz, Erg ̈un, & Ilgaz 2018). Various works (e.g. Göl & Nafalski 2007; Finger et al 2005) highlight the integration of collaboration in studio and project-based learning, but it can also be integrated into more conventional lab-experiment type subjects (Schaf et al 2009).

This work examines the psychological underpinnings, benefits, problems, and practice of collaborative learning with a particular focus on its potential for IT and engineering education at a technical University moving to studio-based learning. The research focus is how collaborative learning is being implemented in IT and engineering and how its use can be improved given the industrial, academic and learning context in the case study University. With the growing push to incorporate these approaches into engineering and IT, it is important that the instructors and students have the tools to best engage in effective collaboration. Selecting these tools may depend on the learning context, the content type and the lecturer’s style.

References

Finger, S., Gelman, D., Fay, A. & Szczerban, M. 2005, ‘Supporting collaborative learning in engineering design’, Proceedings of the Ninth International Conference on Computer Supported Cooperative Work in Design, 2005., IEEE, Coventry, UK, pp. 990-995 Vol. 2.

Göl, Ö. & Nafalski, A. 2007, ‘Collaborative Learning in Engineering Education’, Global Journal of Engineering Education, vol. 11, no. 2, pp. 173–80.

Hathorn, L. & Ingram, A. 2002, ‘Cooperation and Collaboration Using Computer-Mediated Communication’, Journal of Educational Computing Research, vol. 26, no. 26, pp. 325–347.

Kim, J. 2013, ‘Influence of Group Size on Students’ Participation in Online Discussion Forums’, Computers & Education, vol. 62, no. 62, pp. 123–129.

Kurnaz, F. B., Erg ̈un, E., & Ilgaz, H. 2018, ‘Participation in Online Discussion Environments: Is It Really Effective?’, Education and Information Technologies, vol. 23, no. 23, pp. 1719–1736.

Paulus, T. 2005, ‘Collaborative and Cooperative Approaches to Online Group Work: The Impact of Task Type’, Distance Education, vol. 26, no. 26, pp. 111–125.

Schaf, F.M., Müller, D., Bruns, F.W., Pereira, C.E. & Erbe, H.-H. 2009, ‘Collaborative learning and engineering workspaces’, Annual Reviews in Control, vol. 33, no. 2, pp. 246–52.

Stump, G.S., Hilpert, J.C., Husman, J., Chung, W. & Kim, W. 2011, ‘Collaborative Learning in Engineering Students: Gender and Achievement’, Journal of Engineering Education, vol. 100, no. 3, pp. 475–97.

Fast Dictionary-Based Compression for Inverted Indexes

Giulio Ermanno Pibiri1, Matthias Petri2, Alistair Moffat3

1 The University of Pisa, Italy; giulio.pibiri@di.unipi.it 

2 The University of Melbourne, Australia; matthias.petri@gmail.com

3 The University of Melbourne, Australia; ammoffat@unimelb.edu.au

Dictionary-based compression schemes provide fast decoding operation, typically at the expense of reduced compression effectiveness compared to statistical or probability-based approaches. In this work, we apply dictionary-based techniques to the compression of inverted lists, showing that the high degree of regularity that these integer sequences exhibit is a good match for certain types of dictionary methods, and that an important new trade-off balance between compression effectiveness and compression efficiency can be achieved. Our observations are supported by experiments using the document-level inverted index data for two large text collections, and a wide range of other index compression implementations as reference points. Those experiments demonstrate that the gap between efficiency and effectiveness can be substantially narrowed.

This talk will provide an overview of work that will be published in the Proceedings of the 12th ACM International Conference on Web Search and Data Mining, Melbourne, Australia, 11-15 February 2019.

 

 

Using Learning Analytics to Identify Poor Performance and Engagement: Case Study from IT Students

Guy Wood-Bradley1, Sophie McKenzie2, Nick Patterson3, David Tay4, Elicia Lanham5

1 Deakin University, Geelong, Locked Bag 20000, Australia, VIC, 3220, guy.woodbradley@deakin.edu.au 

2 Deakin University, Geelong, Locked Bag 20000, Australia, VIC, 3220, sophie.mckenzie@deakin.edu.au 

3 Deakin University, Geelong, Locked Bag 20000, Australia, VIC, 3220, nick.patterson@deakin.edu.au 

4 Deakin University, Geelong, Locked Bag 20000, Australia, VIC, 3220, david.tay@deakin.edu.au 

5 Deakin University, Geelong, Locked Bag 20000, Australia, VIC, 3220, elicia.lanham@deakin.edu.au 

Learning analytics (LA), using data produced by computer-based educational systems, has been shown to provide an opportunity to impact on the student experience (Peña-Ayala, 2018, Lacave, Molina and Cruz-Lemus, 2018). This poster will explore how LA is being used to better understand the Information Technology (IT) student experience at an Australian University. Current approaches to data analysis, as well as contextual factors, are presented to describe how sense and value can be made from learner data to create impactful results.

LA can be used as a transdisciplinary paradigm to explore areas such as: learner behaviour and performance, social and discourse interaction at learning, prediction of students’ success and attrition rate, and assessment and feedback, learners’ emotions and engagement (Peña-Ayala, 2018). LA is the analysis and reporting of learner related data from diverse fine-grained viewpoints, such as: logins, views, time, and communications. Peña-Ayala (2018) constructed a learning analytics taxonomy to define the current profile and underlying theoretical and contextual factors relevant to LA in higher education. Factors aiding retention include effective orientation and induction, authentic curricula, integrated study skills, and teachers knowing who their students are (Crosling et al. 2009). Higher education organisations experience a percentage of dropouts in their cohorts, reported in the university sector to reach 20% or higher (Strategic Intelligence and Planning Unit, 2017). Existing research also suggest factors impacting progression and student non-completion, such as balancing social and academic activity in university life as well as the integration of students from more diverse backgrounds (Cartney and Rouse 2006).

IT HIGHER EDUCATION STUDENTS – STUDENT PROFILING

With approximately 22,000 students enrolled in an IT course in Victoria in 2016 (Department of Education and Training, 2017), student retention is a key challenge within Australia that varies as institutions strive to keep these loses to a minimum. Notably the results from the Deakin Statistics Summary 2017 (Strategic Intelligence and Planning Unit, 2017) indicates retention rates for higher education domestic students from 2015-2016 with percentages ranging from 75% (Swinburne) to 90% (University of Melbourne).

DEAKIN UNIVERSITY CASE STUDY – CURRENT OUTCOMES

Our results from the subject in IT showed that over 2016 and 2017 there was a decrease in non-participation (XN) from 21 to 16 students, and that those who unenrolled early from the subject stayed relatively the same at 80 students in 2016 and 78 students in 2017. We found that students who received a fail (N) in their initial attempt were largely successful in their second attempt, with smaller amounts of students failing again or withdrawing early. Further we looked into patterns that could result in an N based on student’s assessment results. This led to us looking for a relationship between not submitting assignments and failing a unit by counting non-submitted assignments for N students throughout two study units. Outcomes indicated that failing students did not submit the later subject assignments.

Our Solution

Current work is underway on an application which uses data analytics to review historical activity in the learner data contained on the learning management computer system, which will feed into a model of student data that can inform real-time assessment. This will employ a dedicated algorithm to determine when a student is on the pathway to an XN. In addition the application will focus on early disengagement of students leading to un-enrolment from the unit and identifying students who are likely to leave (WE). When focused on IT students, results from the Student Experience Survey 2016 and 2017 (QILT, 2017) indicate lower in learner engagement (55.1% vs 60.7% nationally) and skills development (69.2% vs 73.8% nationally). Our model should help address these weak points to benefit Computing and Information Systems students.

REFERENCES 

Alejandro Peña-Ayala, (2018), Learning Analytics: A glance of evolution, status, and trends according to a proposed taxonomy, WIREs Data Mining Knowledge Discovery. 2018; 8, 1-29

Crosling, Glenda; Heagney, Margaret and Thomas, Liz. Improving Student Retention in Higher Education: Improving Teaching and Learning [online]. Australian Universities’ Review, The, Vol. 51, No. 2, 2009: 9-18. Availability:<https://search.informit.com.au/documentSummary;dn=159225407205474;res=IELHSS> ISSN: 0818-8068. [cited 14 Mar 18].

Cartney, P. and A. Rouse (2006). “The emotional impact of learning in small groups: highlighting the impact on student progression and retention.” Teaching in Higher Education 11(1): 79-91.

Carmen Lacave, Ana Molina and José Cruz-Lemus (2018), Learning Analytics to identify dropout factors of Computer Science studies through Bayesian networks, Behaviour and Information Technology, DOI: 10.1080/0144929X.2018.1485053

Department of Education and Training. (2017). 2016 All Students.

Mohammed Saqr, Uno Fors & Matti Tedre (2017) How learning analytics can early predict under-achieving students in a blended medical education course, Medical Teacher, 39:7, 757-767, DOI: 10.1080/0142159X.2017.1309376

QILT. (2017). Deakin University – Student Experience – Undergraduate.   Retrieved 27/8/18, from https://www.qilt.edu.au/institutions/list/institution/deakin-university

Strategic Intelligence and Planning Unit. (2017). Deakin Statistics Summary 2017.   Retrieved 27/8/18, from http://www.deakin.edu.au/about-deakin/faculties-and-divisions/administrative-divisions/strategic-intelligence-and-planning-unit