Tag Archives: open science

Data sharing and reuse case study: the Mammographic Image Society database

Uncategorized, , , , ,

The Mammographic Image Society (MIAS) database is a set of mammograms put together in 1992 by a consortium of UK academic institutions and archived on 8mm DAT tape, copies of which were made openly available and posted to applicants for a small administration fee. The mammograms themselves were curated from the UK National Breast Screening Programme, a major screening program that was established in the late 80s offering routine screening every three years to women aged between 50-64.

The motivations for creating the database were to make a practical contribution to computer vision research – which sought to improve the ability of computers to interpret images – and to encourage the creation of more extensive datasets. In the peer-reviewed paper bundled with the dataset, the researchers note that “a common database is a positive step towards achieving consistency in performance comparison and testing of algorithms”.

Due to increased demand, the MIAS database was made available online via third parties, albeit in a lower resolution than the original. Despite no longer working in this area of research, the lead author, John Suckling – now Director of Research in the Department of Psychiatry, part of Cambridge Neuroscience –  started receiving emails asking for access to the images at the original resolution. This led him to dig out the original 8mm DAT tapes with the intention of making the images available openly in a higher resolution. The tapes were sent to the University Information Service (UIS), who were able to access the original 8mm tape and download higher resolution versions of the images. The images were subsequently deposited in Apollo and made available under a CC BY license, meaning researchers are permitted to reuse them for further research as long as appropriate credit is given. This is the most commonly used license for open datasets and is recommended by the majority of research funding agencies.

Motivations for sharing the MIAS database openly

The MIAS database was created with open access in mind from the outset. When asked whether he had any reservations about sharing the database openly, the lead author John Suckling noted:

“There are two broad categories of data sharing; data acquired for an original purpose that is later shared for secondary use; data acquired primarily for sharing. This dataset is an example of the latter. Sharing data for secondary use is potentially more problematic especially in consortia where there are a number of continuing interests in using the data locally. However, most datasets are (or should be) superseded, and then value can only be extracted if they are combined to create something greater than the sum of the parts. Here, careful drafting of acknowledgement text can be helpful in ensuring proper credit is given to all contributors.”

This distinction – between data acquired for an original purpose that is later shared for secondary use and data acquired primarily for sharing – is one that is important and often overlooked. The true value of some data can only be fully realised if openly shared. In such cases, as Suckling notes, sufficient documentation can help ensure the original researchers are given credit where it is due, as well as ensuring it can be reused effectively. This is also made possible by depositing the data on an institutional repository such as Apollo, where it will be given a DOI and its reuse will be easier to track.

Impact of the MIAS database

As of August 2020, the MIAS database has received over 5500 downloads across 27 different countries, including some developing countries where breast cancer survival rates are lower. Google Scholar currently reports over 1500 citations for the accompanying article as well as 23 citations for the dataset itself. A review of a sample of the 1500 citations revealed that many were examples of the data being reused rather than simply citations of the article. Additionally, a systematic review published in 2018 cited the MIAS database as one of the most widely used for applying breast cancer classification methods in computer aided diagnosis using machine learning, and a benchmarking review of databases used in mammogram research identified it as the most easily accessible mammographic image database. The reasons cited for this included the quality of the images, the wide coverage of types of abnormalities, and the supporting data which provides the specific locations of the abnormalities in each image.

The high impact of the MIAS database is something Suckling credits to the open, unrestricted access to the database, which has been the case since it was first created. When asked whether he has benefited from this personally, Suckling stated “Direct benefits have only been the citations of the primary article (on which I am first author). However, considerable efforts were made by a large number of early-career researchers using complex technologies and digital infrastructure that was in its infancy, and it is extremely gratifying to know that this work has had such an impact for such a large number of scientists.”. Given that the database continues to be widely cited and has been downloaded from Apollo 1358 times since January 2020, it is still clearly the case that the MIAS database is having a wide impact.

The MIAS Database Reused

As mentioned above, the MIAS database has been widely reused by researchers working in the field of medical image analysis. While originally intended for use in computer vision research, one of the main ways in which the dataset has been used is in the area of computer aided diagnosis (CAD), for which researchers have used the mammographic images to experiment with and train deep learning algorithms. CAD aims to augment manual inspection of medical images by medical professionals in order to increase the probability of making an accurate diagnosis.

A 2019 review of recent developments in medical image analysis identified lack of good quality data as one of the main barriers researchers in this area face. Not only is good quality data a necessity but it must also be well documented as this review also identified inappropriately annotated datasets as a core challenge in CAD. The MIAS database is accompanied by a peer-reviewed paper explaining its creation and content as well as a read me PDF which explains the file naming convention used for the images as well as the annotations used to indicate the presence of any abnormalities and classify them based on their severity. The presence of this extensive documentation combined with it having been openly available from the outset could explain why the database continues to be so widely used.

Reuse example: Applying Deep Learning for the Detection of Abnormalities in Mammograms

This research, published in 2019 in Information Science and Applications, looked at improving some of the current methods used in CAD and attempted to address some inherent shortcomings and increase the competency level of deep learning models when it comes the minimisation of false positives when applying CAD to mammographic imaging. The researchers used the MIAS database alongside another larger dataset in order to evaluate the performance of two existing convolutional neural networks (CNN), which are deep learning models used specifically for classifying images. Using these datasets, they were able to demonstrate that versions of two prominent CNNs were able to detect and classify the severity of abnormalities on the mammographic images with a high degree of accuracy.

While the researchers were able to make good use of the MIAS database to carry out their experiments, due to the inclusion of appropriate documentation and labelling, they do note that since it is a relatively small dataset it is not possible to rule out “overfitting”, where a deep learning model is highly accurate on the data used to train the model, but may not generalise well to other datasets. This highlights the importance of making such data openly available as it is only possible to improve the accuracy of CAD if sufficient data is available for researchers to carry out further experiments and improve the accuracy of their models. ­

Reuse example: Computer aided diagnosis system for automatic two stages classification of breast mass in digital mammogram images

This research, published in 2019 in Biomedical Engineering: Applications, Basis and Communications, used the MIAS database along with the Breast Cancer Digital Repository to test a CAD system based on a probabilistic neural network – a machine learning model that predicts the probability distribution of a given outcome –  developed to automate classification of breast masses on mammographic images. Unlike previously developed models, their model was able to segment and then carry out a two-stage classification of breast masses. This meant that rather than classifying masses into either benign or malignant, they were able to develop a system which carried out a more fine-grained classification consisting of seven different categories. Combining the two different databases allowed for an increased confidence level in the results gained from their model, again raising the importance of the open sharing of mammographic image datasets. After testing their model on images from these databases, they were able to demonstrate a significantly higher level of accuracy at detecting abnormalities than had been demonstrated by two similar models used for evaluation. On images from the MIAS Database and Breast Cancer Digital Repository their model was able to detect abnormalities with an accuracy of 99.8% and 97.08%, respectively. This was also accompanied by increased sensitivity (ability to correctly classify true positives) and specificity (ability to correctly classify false negatives).

Conclusion

Many areas of research can only move forward if sufficient data is available and if it is shared openly. This, as we have seen, is particularly true in medical imaging where despite datasets such as the MIAS database being openly available, there is a data deficiency which needs to be addressed in order to improve the accuracy of the models used in computer-aided diagnosis. The MIAS database is a clear example of a dataset that has enabled an important area of research to move forward by enabling researchers to carry out experiments and improve the accuracy of deep learning models developed for computer-aided diagnosis in medical imaging. The sharing and reuse of the MIAS database provides an excellent model for how and why future researchers should make their data openly available.

Published 20th August 2020
Written by Dominic Dixon

CCBY icon

The case for Open Research: solutions?

Uncategorized, , , , , , ,

This series arguing the case for Open Research has to date looked at some of the issues in scholarly communication today. Hyperauthorship, HARKing, the reproducibility crisis, a surge in retractions all stem from the requirement that researchers publish in high impact journals. The series has also looked at the invalidity of the impact factor and issues with peer review.

This series is one of an increasing cacophony of calls to move away from this method of rewarding researchers. Richard Smith noted in a recent BMJ blog criticising the current publication in journal system: “The whole outdated enterprise is kept alive for one main reason: the fact that employers and funders of researchers assess researchers primarily by where they publish. It’s extraordinary to me and many others that the employers, mainly universities, outsource such an important function to an arbitrary and corrupt system.”

Universities need to open research to ensure academic integrity and adjust to support modern collaboration and scholarship tools, and begin rewarding people who have engaged in certain types of process rather than relying on traditional assessment schemes. This was the thrust of a talk in October last year”Openness, integrity & supporting researchers“. If nothing else, this approach makes ‘nightmare scenarios’ less likely. As Prof Tom Cochrane said in the talk, the last thing an institution needs is to be on the front page because of a big fraud case. 

What would happen if we started valuing and rewarding other parts of the research process? This final blog in the series looks at opening up research to increase transparency. The argument suggests we need to move beyond rewarding only the journal article – and not only other research outputs, such as data sets but research productivity itself.

So, let’s look at how opening up research can address some of the issues raised in this series.

Rewarding study inception

In his presentation about HARKing (Hypothesising After the Results are Known) at FORCE2016 Eric Turner, Associate Professor OHSU suggested that what matters is the scientific question and methodological rigour. We should be emphasising not the study completion but study inception before we can be biased by the results.  It is already a requirement to post results of industry sponsored research in ClinicalTrials.gov – a registry and results database of publicly and privately supported clinical studies of human participants conducted around the world. Turner argues we should be using it to see the existence of studies.  He suggested reviews of protocols should happen without the results (but not include the methods section because this is written after the results are known).

There are some attempts to do this already. In 2013 Registered Reports was launched: “The philosophy of this approach is as old as the scientific method itself: If our aim is to advance knowledge then editorial decisions must be based on the rigour of the experimental design and likely replicability of the findings – and never on how the results looked in the end.” The proposal and process is described here. The guidelines for reviewers and authors are here, including the requirement to “upload their raw data and laboratory log to a free and publicly accessible file-sharing service.”

This approach has been met with praise by a group of scientists with positions on more than 100 journal editorial boards, who are “calling for all empirical journals in the life sciences – including those journals that we serve – to offer pre-registered articles at the earliest opportunity”. The signatories noted “The aim here isn’t to punish the academic community for playing the game that we created; rather, we seek to change the rules of the game itself.” And that really is the crux of the argument. We need to move away from the one point of reward.

Getting data out there

There is definite movement towards opening research. In the UK there is now a requirement from most funders that the data underpinning research publications are made available. Down under, the Research Data Australia project is a register of data from over 100 institutions, providing a single point to search, find and reuse data. The European Union has an Open Data Portal.

Resistance to sharing data amongst the research community is often due to the idea that if data is released with the first publication then there is a risk that the researcher will be ‘scooped’ before they can get those all-important journal articles out. In response to this query during a discussion with the EPSRC it was pointed out that the RCUK Common Principles state that those who undertake Research Council funded work may be entitled to a limited period of privileged use of the data they have collected to enable them to publish the results of their research. However, the length of this period varies by research discipline.

If the publication of data itself were rewarded as a ‘research output’ (which of course is what it is), then the issue of being scooped becomes moot. There have been small steps towards this goal, such as a standard method of citing data.

A new publication option is Sciencematters, which allows researchers to submit observations which are subjected to triple-blind peer review, so that the data is evaluated solely on its merits, rather than on the researcher’s name or organisation. As they indicate “Standard data, orphan data, negative data, confirmatory data and contradictory data are all published. What emerges is an honest view of the science that is done, rather than just the science that sells a story”.

Despite the benefits of having data available there are some vocal objectors to the idea of sharing data. In January this year a scathing editorial in the New England Journal of Medicine suggested that researchers who used other people’s data were ‘research parasites’. Unsurprisingly this position raised a small storm of protest (an example is here). This was so sustained that four days later a clarification was issued, which did not include the word ‘parasites’.

Evaluating & rewarding data

Ironically, one benefit of sharing data could be an improvement to the quality of the data itself. A 2011 study into why some researchers were reluctant to share their data found this to be associated with weaker evidence (against the null hypothesis of no effect) and a higher prevalence of apparent errors in the reporting of statistical results. The unwillingness to share data was particularly clear when reporting errors had a bearing on statistical significance.

Professor Marcus Munafo in his presentation at the Research Libraries UK conference held earlier this year suggested that we need to introduce quality control methods implicitly into our daily practice. Open data is a very good step in that direction. There is evidence that researchers who know their data is going to be made open are more thorough in their checking of it. Maybe it is time for an update in the way we do science – we have statistical software that can run hundreds of analysis, and we can do text and data mining of lots of papers. We need to build in new processes and systems that refine science and think about new ways of rewarding science.

So should researchers be rewarded simply for making their data available? Probably not, some kind of evaluation is necessary. In a public discussion about data sharing held at Cambridge University last year, there was the suggestion that rather than having the formal peer review of data, it would be better to have an evaluation structure based on the re-use of data – for example, valuing data which was downloadable, well-labelled and re-usable.

Need to publish null results

Generally, this series looking at the case for Open Research has argued that the big problem is the only thing that ‘counts’ is publication in high impact journals. So what happens to all the results that don’t ‘find’ anything?

Most null results are never published with a study in 2014 finding that of 221 sociological studies conducted between 2002 and 2012, only 48% of the completed studies had been published. This is a problem because not only is the scientific record inaccurate, it means  the publication bias “may cause others to waste time repeating the work, or conceal failed attempts to replicate published research”.

But it is not just the academic reward system that is preventing the widespread publication of null results – the interference of commercial interests on the publication record is another factor. A recent study looked into the issue of publication agreements – and whether a research group had signed one prior to conducting randomised clinical trials for a commercial entity. The research found that  70% of protocols mentioned an agreement on publication rights between industry and academic investigators; in 86% of those agreements, industry retained the right to disapprove or at least review manuscripts before publication. Even more concerning was  that journal articles seldom report on publication agreements, and, if they do, statements can be discrepant with the trial protocol.

There are serious issues with the research record due to selected results and selected publication which would be ameliorated by the requirement to publish all results – including null results.

There are some attempts to address this issue. Since June 2002 the Journal of Articles in Support of the Null Hypothesis has been published bi-annually. The World Health Organisation has a Statement on the Public Disclosure of Clinical Trial Results, saying: “Negative and inconclusive as well as positive results must be published or otherwise made publicly available”. A project launched in February last year by PLOS ONE is a collection focusing on negative, null and inconclusive results. The Missing Pieces collection had 20 articles in it as of today.

In January this year there were reports that a group of ten editors of management, organisational behaviour and work psychology research had pledged they would publish the results of well-conceived, designed, and conducted research even if the result was null.  The way this will work is the paper is presented without results or discussion first and it is assessed on theory, methodology, measurement information, and analysis plan.

Movement away from using the impact factor

As discussed in the first of this series of blogs ‘The mis-measurement problem‘, we have an obsession with high impact journals. These blogs have been timely, falling as they have within what seems to be a plethora of similarly focused commentary. An example is a recent Nature news story by Mario Biagioli, who argued the focus on impact of published research has created new opportunities for misconduct and fraudsters. The piece concludes that “The audit culture of universities — their love affair with metrics, impact factors, citation statistics and rankings — does not just incentivize this new form of bad behaviour. It enables it.”

In recent discussion amongst the Scholarly Communication community about this mis-measurement the suggestion that we can address the problem by limiting the number of articles that can be submitted for promotion was raised. This ideally reduces the volume of papers produced overall, or so the thinking goes. Harvard Medical School and the Computing Research Association “Best Practices Memo” were cited as examples by different people.

This is also the approach that has been taken by the Research Excellence Framework in the UK – researchers put forward their best four works from the previous period (typically about five years). But it does not prevent poor practice. Researchers are constantly evaluated for all manner of reasons. Promotion, competitive grants, tenure, admittance to fellowships are just a few of the many environments a researcher’s publication history will be considered.

Are altmetrics a solution? There is a risk that any alternative indicator becomes an end in itself. The European Commission now has an Open Science Policy Platform, which, amongst other activities has recently established an expert group to advise on the role of metrics and altmetrics in the development of its agenda for open science and research.

Peer review experiments

Open peer review is where peer review reports identify the reviewers and are published with the papers.  One of the more recent publishers to use this method of review is the University of California Press’ open access mega journal called Collabra, launched last year. In an interview published by Richard Poynder, UC Press Director Alison Mudditt notes that there are many people who would like to see more transparency in the peer review process. There is some evidence to show that identifying reviewers results in more courteous reviews.

PLOS One publishes work after an editorial review process which does not include potentially subjective assessments of significance or scope to focus on technical, ethical and scientific rigor. Once an article is published readers are able to comment on the work in an open fashion.

One solution could be that used by CUP journal JFM Rapids, which has a ‘fast-track’ section of the journal offering fast publication for short, high-quality papers. This also operates a policy whereby no paper is reviewed twice, thus authors must ensure that their paper is as strong as possible in the first instance. The benefit is it offers a fast turnaround time while reducing reviewer fatigue.

There are calls for post publication peer review, although some attempts to do this have been unsuccessful, there are arguments that it is simply a matter of time – particularly if reviewers are incentivised. One publisher that uses this system is the platform F1000Research which publishes work immediately and invites open post-publication review. And, just recently, Wellcome Open Research was launched using services developed by F1000Research. It will make research outputs available faster and in ways that support reproducibility and transparency. It uses an open access model of immediate publication followed by transparent, invited peer review and inclusion of supporting data.

Open ways of conducting research

All of these initiatives demonstrate a definite movement towards an open way of doing research by addressing aspects of the research and publication process. But there are some research groups that are taking a holistic approach to open research.

Marcus Munafo published last month a description of the experience the UK Center for Tobacco and Alcohol Studies and the MRC Integrative Epidemiology Unit at the University of Bristol over the past few years of attempting to work within an Open Science Model focused on three core areas:  study protocols, data, and publications.

Another example is the Open Source Malaria project which includes researchers and students using open online laboratory notebooks from around the world including Australia, Europe and North America. Experimental data is posted online each day, enabling instant sharing and the ability to build on others’ findings in almost real time. Indeed, according to their site ‘anyone can contribute’. They have just announced that undergraduate classes are synthesising molecules for the project. This example fulfils all of the five basic principles of open research suggested here.

The Netherlands Organisation for Scientific Research (NWO) has just announced that it is making 3 million euros available for a Replication Studies pilot programme. The pilot will concentrate on the replication of social sciences, health research and healthcare innovation studies that have a large impact on science, government policy or the public debate. The intention after this study will be to “include replication research in an effective manner in all of its research programmes”.

A review of literature published this week has demonstrated that open research is associated with increases in citations, media attention, potential collaborators, job opportunities and funding opportunities. These findings are evidence, the authors say,  “that open research practices bring significant benefits to researchers relative to more traditional closed practices”.

This series has been arguing that we should move to Open Research as a way of changing the reward system that bastardises so much of the scientific endeavour. However there may be other benefits according to a recently published opinion piece which argues that Open Science can serve a different purpose to “help improve the lot of individual working scientists”.

Conclusion

There are clearly defined problems within the research process that in the main stem from the need to publish in  high impact journals. Throughout this blog there are multiple examples of initiatives and attempts to provide alternative ways of working and publishing.

However, all of this effort will only succeed if those doing the assessing change the rules of the game. This is tricky. Often the people who have succeeded have some investment in the status quo remaining. We need strong and bold leadership to move us out of this mess and towards a more robust and fairer future. I will finish with a quote that has been attributed to Mark Twain, Einstein and Henry Ford. “If you always do what you’ve always done, you’ll always get what you’ve always got”. It really is up to us.

Published 2 August 2016
Written by Dr Danny Kingsley
Creative Commons License