SPIE 2015 Talks Big [data]

This past weekend, I had the opportunity to attend the SPIE Medical Imaging conference held in Orlando, FL. My visit was cut short to a single full day, but it was enough to learn a great deal.

The meeting is divided into different tracks, each a themed conference with a keynote speaker, paper presentations, and a workshop. Attendees are free to switch from room to room to attend different topics. These are generally engineers, physicists, computer scientists, as well as industry leaders in imaging technology like IBM and Siemens. I spent most of my day in the PACS and Medical Informatics (9418) conference.

SPIE Medical Imaging conference is held in Orlando FL this year

SPIE Medical Imaging conference is held in Orlando FL this year

Keynote Speaker

Dr. Eliot Siegel from University of Maryland delivered the keynote speech at the start of the day for PACS and Medical Informatics giving an overview of challenges for “big data” in medical imaging. Computer-aided detection was mentioned as an area of technological advance where not enough practices are adapting despite clinical benefits. He urged continued efforts in big data, using it to curb the challenges associated with the ever-increasing volume of radiologic studies. Dr. Siegel spent the bottom half of the keynote speech using a case example of a patient with pulmonary nodules as an example of how big data can be used to answer his clinical questions, where no existing solution can identify the patient’s personal, exact risk of malignancy.

Advances in Big Data

Dr. Jianguo Zhang, one of the chairs for the conference then presented the challenges of big data, and how they can be tackled. He gave an overview of what we could expect to see for the rest of the day in this conference, highlighting research projects using grid-based, cloud-based, solutions. He also mentioned the advent of semantic search engines, 3D imaging solutions, and segmentation.

Question and Answer

In the question and answer session that ensued, conference attendees raised several interesting points, which are outlined as follows:

  1. Standardization – There is no standard format for big data. Data can be normalized to allow sharing, but in the absence of standardized schema or data format, this task remains difficult. Efforts made by Google and NCI have not made great strides so far.

  2. Beyond Imaging – “Big Data” is not just an imaging problem. Mapping of data by scale in radiology and in pathology are similar and synergistic. Multiparametric data in radiology and pathology makes them incredibly difficult to combine. However, the key point is that big data exists beyond the genome, which was where the term first gained popularity.

  3. Privacy – Some data – like genomic sequences – are inherently identifying (the genome is, after all, the ultimate identification of an individual). A big challenge will be to protect patient privacy in the setting of big data when the data itself cannot be de-identified.

  4. Security – Despite our best efforts, sometimes the data itself can also be processed to de-anonymize a patient by interested parties using clever processing tools. Dr. Siegel noted that patients themselves are often less concerned than the doctors. When asked what they do with their x-ray images, patients often times say they simply share with family and friends on Facebook!

  5. Record-keeping – Electronic medical records are highly heterogeneous. At the same time they rely on humans to update. How can we ensure that every patient’s record is updated with the same level of meticulousness that one would pay to charts used in clinical trials? Just as importantly, how can we do this without increasing the underlying cost?

  6. Cloud Challenges – Cloud-based computing the speed will never match the speed of local storage, so how can a radiology practice move data from clinical practice to storage and vice versa, with reasonable turnaround time?

Big Data

A lot of incredible research projects were presented as part of this conference. I cannot even pretend to understand all of the great work that is being done but will attempt to briefly summarize those I attended.

One project called DicomSpider was presented by Lars Lindsk√∂ld, created to study how different cultures can combine data and use machine generated metadata for this purpose. The researchers noted that “standards of DICOM do not have a data model for enterprise sharing.” Therefore, combining data from multiple institutions was a difficult task.

Baldur van Lew presented a project called EYR-Garlic (funded by an organization called VANPIRE, no less) combines genetic/genomic information with structural and physiologic information via light-path circuits connecting among research or educational networks, isolated from the internet.

Nina Bougatf, part of a radiation oncology research group from Universitäts Klinikum Heidelberg achieved automation of heterogeneous and distributed data for radiation oncology. The project provided standardized access to research database, imaging, and analytic tools for retrospective clinical trials.

Phillipe Journeau presented a paper on the nature of research communities using concepts of medical imaging (semiotics). I believe the details of the research is above my ability to understand them… but the work is focused on using relational maps to study the boundaries and features of research fields. Journeau used publications around laser research as an example.

Neuroimaging Research

Several abstracts I had the opportunity to listen to appear to share a theme of neuroradiology. I am unsure whether this was a sign of the type of SPIE research in vogue or simply by chance.

Kevin Ma presented a project on data analysis for multiple sclerosis by optimizing archive and analysis for data using a disease-centric approach. The tool stores clinical data, laboratory data, imaging data, and post-processing.

Ximing Wang studied stroke lesions using digital templates and volume of stroke lesions in a web-based imaging system.

Applying clever software modifications to existing PACS workflow, Justin Sensenery incorporated research images into the clinical PACS system for traumatic brain injury, integrating research and clinical workflow using the same interface. Unfortunately, the presenter admitted that such a tool was unlikely to be directly transferable to a different institution without extensive modification.

Workflow Projects

Web Based PACS

Three presentations were made by the creators of netDicom. Ashesh Parikh and Nihal Mehta together spoke about the implementation of a cloud-based storage for DICOM images. The services they chose were Microsoft Azure and Amazon Cloud Services. They then developed a web-based interface using standard languages in such as HTML5, AJAX, and JavaScript to enable viewing. It also uses FHIR/HL7 data feeds for real-time update of data.


Leticia Rittner presented a web-based platform for medical imaging research called Adessowiki, as a document writing and software platform. Use of a wiki allows easier modification, integration between software developers without necessity off installing hardware.

Cancer Collaboration

Sebastian Roberto Tarando worked as part of a group creating an online collaborative environment, as cancer follow up require extensive collaboration. Challenges need to be solved include preserving lossless image data quality and latency problems. Primary challenges for implementing this tool include latency and ability to compress the data.

Radiology for Non-Radiologists

I was fortunate enough to work with Drs. Steve Horii and Tessa Cook to deliver the workshop at the end of the day. Despite being scheduled to start at 5:45pm and slated to run for 2 hours, the workshop was well-attended. Together, we described the traditional job description of a radiologist, highlighted the complexity of workflow, and attempted to build a framework to think about the challenges ahead.

Just as most radiologists cannot describe the inner working of the technology despite using them every day, many engineers can benefit from a closer understanding of radiology workflow. A common theme that emerged in the ensuing Q&A was the general puzzling notion of why good science gets left at the bench and never become adapted in clinical work. The general sense is that a product, no matter how thoroughly researched the supporting science, must find a fit in radiology workflow have hope of being adapted in practice.


In an 1997 article in the Harvard Business Review, Dorothy Leonard and Susan Straus state “To innovate successfully, you must hire, work with, and promote people who are unlike you.” Likewise, to be a well-balanced radiologist, it is worth spending time learning from non-radiologists whose work powers our profession. SPIE is an excellent conference for radiologists because of its heavy focus on the physics, computation, and engineering all moving towards the singular goal of propelling medical imaging. In some ways it is identical to the goals of academic radiologists, but in many ways it is dramatically different. And in sorting out these differences, radiologists and engineers become better at their own work.

Howard Chen on GithubHoward Chen on LinkedinHoward Chen on Wordpress
Howard Chen
Vice Chair for Artificial Intelligence at Cleveland Clinic Diagnostics Institute
Howard is passionate about making diagnostic tests more accurate, expedient, and affordable through disciplined implementation of advanced technology. He previously served as Chief Informatics Officer for Imaging, where he led teams deploying and unifying radiology applications and AI in a multi-state, multi-hospital environment. Blog opinions are his own and in no way reflect those of the employer.

Leave a Reply