So far three articles I have contributed to have appeared on the Ellumen website. In case you missed it below is a brief summary of these articles. Be sure to check them out for additional details.

Could AI Be the Radiologist’s Best Friend?

The Could AI Be the Radiologist’s Best Friend? article published on February 17, 2021. The article discusses how AI has the potential to alleviate the demand on radiologists by doing preliminary evaluations on medical images and organizing imaging workflows to improve efficiency. The article mentions the five most common use cases of AI in radiology today: 1) optimizing workflow for productivity, 2) tagging images so critical patients are the first reviewed, 3) automating part of the image analysis, 4) enhancing imaging quality, and 5) providing decision support, and presents an excellent graphic to accompany this to improve understanding. The article further discusses how by positioning AI technology as a useful and supplemental tool, radiologists and clinicians can reap the benefits while their confidence in using AI grows and skepticism fades.

AI for Medical Imaging Research: A Guide to Accessible Tools and Resources

The AI for Medical Imaging Research: A Guide to Accessible Tools and Resources article published on March 22, 2021. The article discusses numerous tools and resources that currently exist that researchers can use to help develop AI algorithms for medical imaging. The article also provides an extensive list of medical imaging datasets with high quality images and annotations that already exist. Further, it is discussed how it is hoped with improved awareness of the numerous tools and data sources available to AI researchers today, participation in medical imaging research and progress to accelerate AI in medical imaging transformation will be made.

AI Tools in Triage Lead to Faster Diagnoses

The AI Tools in Triage Lead to Faster Diagnoses article published on May 4, 2021. This article discusses utilizing AI as a triage mechanism and in support of more efficient workflows for medical imaging diagnosis. It is known that patient outcomes are directly correlated with speed and in many cases patient care is extremely time sensitive. The article discusses how AI can be be utilized to reduce the time required for an MRI scan from one hour to 15 minutes and by doing so reduce the noise in images and allow more patients to be scanned by the same MRI machine each day. The article also presents details on how a deep learning neural network can be trained using labeled images of diseases and normal conditions present and shows a graphic to further understanding. The neural network can be leveraged to provide radiologists with a screening tool before they look at a series of images, allowing them to more quickly move through the series and form an opinion on a diagnosis.

Future Articles for Ellumen’s Blog Series on AI Innovation in Medical Imaging

It is believed that AI in medical imaging can lead to better outcomes for patients. Radiologists who recognize the importance of AI’s medical imaging transformation will lead to improvements in patient care and accuracy of a diagnosis. Be sure to keep an eye out for new forthcoming articles on the Ellumen website related to AI innovation in medical imaging and feel free to reach out to experts at Ellumen to help explore the potential of AI to solve medical imaging research needs.

The post Introducing Ellumen’s Blog Series on AI Innovation in Medical Imaging and Roundup of Three Recent Articles appeared first on Todd McCollough's Website.

This year included live meeting sessions running from 8 a.m. to 6 p.m. on November 29 to December 5th. If you were unable to attend the live programming as it happened, it converted to on demand sessions thereafter. One advantage of the live programming sessions was the ability to chat with other attendees while it was occurring. However, not all of the live programming sessions were in fact live even though they broadcasted originally live. Instead many of these live sessions occurred without the normal moderator present and were prerecorded. The meeting also included on-demand content available 24/7 during that time period. This year many exhibitors had virtual booths for exhibition and there was also virtual networking available. For those who paid for a premium registration, there was also the benefit of extended on-demand access to most meeting content until April 30, 2021. In addition, case of the day and digital posters were available at RSNA 2020. Navigation presented to users upon login to RSNA 2020 was as below:

Outside of virtual booths, this year included various ways to get the latest on industry developments including two theaters: an Innovation Theater and AI (artificial intelligence) Theater. This was in addition to Lunch & Learns, Featured Demonstrations, Pre Show Presentations, and Roundtable Discussions. There was also an Imaging AI In Practice Demonstration including four videos on the following topics: 1) Introduction, 2) Work with AI, 3) Evolve with AI, and 4) Save Time with AI. These Imaging AI In Practice Demonstration videos included helpful flow-charts to describe how the AI Orchestrator and AI Algorithms integrate into the radiology workflow:

There was certainly a large amount of content focused around AI at RSNA 2020. I attended a session titled Artificial Intelligence: Beyond Interpretive Considerations that consisted of four separate talks. One talk in this session discussed how there was minimal merger and acquisition activity in radiology AI in 2020 and there were six companies involved with radiology AI that previously attended RSNA in prior years that appear to no longer be active. Another talk discussed Generative Adversarial Networks and their potential to create synthetic data in radiology. An example was shown on using a StyleGan2 (a type of GAN) to create synthetic chest radiographs. Another talk discussed the liability risks in using AI in medicine. Those potentially at risk include radiologists, healthcare systems, and even AI developers. The presenter discussed how due to lack of meaningful case law to date a lot remains unknown. I also attended another session titled Creating Publicly Accessible Radiology Imaging Resources for Machine Learning and AI. One talk in this session discussed using the The Cancer Imaging Archive (TCIA) as a dataset for AI training. Another talk mentioned some other datasets including the Alzheimer’s Disease Neuroimaging Initiative (ADNI), Medical Information Mart for Intensive Care (MIMIC), National Biomedical Imaging Archive (NBIA), the Lung Image Database Consortium (LIDC), and datasets from Stanford University’s Center for Artificial Intelligence in Medicine and Imaging. In addition, discussion was made of efforts from developing the Medical Imaging and Data Resource Center (MIDRC), an open-source database with medical images from COVID-19 patients, being collaborated on across more than 20 organizations in the U.S.

This year there appeared to be less focus on 3D printing. Although one live session titled Medical 3D Printing Regulatory and Quality Considerations was available. One talk in this session discussed sterilization for 3D printed devices. There was discussion made of sterilization approaches for 3D printed devices both in the hospital (steam, hydrogen peroxide, and ethylene oxide) and outside the hospital (gamma radiation, e-beam, hydrogen peroxide, and ethylene oxide). There were also at least two on-demand sessions available in the Innovation Theater from two industry leaders in 3D printing: 1) Extending Access to Extended Reality from Materialise and 2) Learnings from the Field: Clinical Care and Device Development in the COVID Era from Formlabs Medical.

The sponsors of RSNA 2020 included those below:

If you are interested learning more about past in person RSNA meetings at McCormick Place, I attended the RSNA annual meeting last year in 2019, and the prior two years in 2018 and 2017. You can find more information and numerous photographs at http://www.toddmccollough.com/radiological-society-of-north-america-rsna-meeting-in-chicago-il-in-2019-at-mccormick-place, http://www.toddmccollough.com/radiological-society-of-north-america-rsna-meeting-in-chicago-il-in-2018-at-mccormick-place/, and http://www.toddmccollough.com/radiological-society-north-america-rsna-chicago-il-2017-mccormick-place/.

The post Virtual Radiological Society of North America (RSNA) Meeting in 2020 appeared first on Todd McCollough's Website.

This year also expanded focus on 3D Printing and Advanced Visualization with an expanded Showcase and Theater offering daily presentations on the latest research and innovations in 3D printing for medical applications. I was able to attend a presentation covering Category III CPT Codes for 3D Printing of Anatomic Models and Guides, Scripting for Segmentation, 3D Printing to Support Research, and Leveraging 3D Printing for Surgical Simulation. It was quite interesting to hear more about the Category III CPT Codes for 3D Printing, which includes 0559T, 0560T, 0561T, and 0562T that went into effect in July, 2019. This should allow for greater adoption by physicians and medical centers. Even so, for those utilizing 3D printing, it was encouraged by the presenter of the CPT code talk to sign up for the RSNA-ACR 3D Printing Registry to help support a future category I CPT code.

As usual there were numerous posters and presentations. Also as usual, there were many exhibitors with medical imaging devices ready to provide demonstrations of their latest technology. New exhibitors this year included Amazon Web Services (AWS) and Medical IP. I was able to attend a few educational courses and scientific sessions. In particular I attended the Artificial Intelligence: Cutting Edge Artificial Intelligence session and Creating Publicly Accessible Radiology Imaging Resources for Machine Learning and AI sessions. In the former session mentioned above, an interesting talk titled Defacing Neuroimages discussed image de-identification using a two-step deep learning model for head CTs and brain MRIs. In the former session, I also was intrigued by a talk titled Automated Detection of Vertebral Fractures in CT Using 3D Convolutional Neural Networks that discussed automatically detecting vertebral fractures in CT images of the spine using a learning method with 3D features. The latter session featured several talks discussing practical challenges with data preparation including image pre-processing steps, techniques for creating ground truth labeling, and statistical approaches to create training and testing data sets.

Below are some of the pictures I took while at the RSNA annual meeting in 2019, in Chicago, IL.

I attended the RSNA annual meeting last year in 2018 and the prior year in 2017, where you can find more information and photos at http://www.toddmccollough.com/radiological-society-of-north-america-rsna-meeting-in-chicago-il-in-2018-at-mccormick-place/ and http://www.toddmccollough.com/radiological-society-north-america-rsna-chicago-il-2017-mccormick-place/.

The post Radiological Society of North America (RSNA) Meeting in Chicago, IL, in 2019, at McCormick Place appeared first on Todd McCollough's Website.

This year brought back the popular deep learning classroom presented by the NVIDIA Deep Learning Institute (DLI) designed for attendees to engage with deep learning tools, write algorithms and improve their understanding of deep learning technology. In one session, called Introduction to Deep Learning, attendees used convolutional neural networks (CNNs) along with a MedNIST data set that consists of 1,000 images each from 5 different categories: Chest X-ray, hand X-ray, Head CT, Chest CT, Abdomen CT, and Breast MRI. The task was for the attendees to identify the image type. Another session focused on 3D segmentation of Brain MR using deep learning methods for segmentation, particularly V-nets.

This year also brought back the machine learning showcase which allowed for the opportunity to network with nearly 80 companies on the forefront of the developments in machine learning and artificial intelligence. This year introduced a new showcase called the 3D printing & advanced visualization showcase which focused on groundbreaking technology in 3D printing, virtual reality and augmented reality. Another new feature this year was a Recruiters Row which allow for attendees to connect with organizations offering career opportunities. Like last year there was also a start-up showcase that featured emerging companies bringing innovations in medical imaging.

I was able to attend a few educational courses and scientific sessions. In particular I attended a session titled Image Processing in Imaging and Radiation Therapy and another session titled Deep Learning in Radiology: How Do We Do It? In the former session I was intrigued by the talk from ImBio which trained a CNN to create quality ventricle segmentations with only 43 scans in the training dataset and used data augmentations to improve the performance on the test dataset and another talk from researchers at the University of Chicago to classify chest radiographs as anteroposterior or posteroanterior. The latter session as indicated above I attended featured insights into deep learning in radiology at The Ohio State University, Stanford University, and the Mayo Clinic Rochester.

Below are some of the pictures I took while at the RSNA annual meeting in 2018, in Chicago, IL.

I also attended last years RSNA annual meeting in 2017 which you can find more information and photos at here http://www.toddmccollough.com/radiological-society-north-america-rsna-chicago-il-2017-mccormick-place/.

The post Radiological Society of North America (RSNA) Meeting in Chicago, IL, in 2018, at McCormick Place appeared first on Todd McCollough's Website.

The first and third patents titled “Dielectric Encoding of Medical Images” resulted from wanting a way to allow for doctors to easily read and understand images produced using electromagnetics represented in dielectric values. To accomplish this I worked with the chief executive officer (CEO) of Ellumen Inc. to explore the microwave imaging modality while also allowing for easy adaptability by doctors and hospitals. I researched the modality, developed algorithms, and developed programs to convert medical images in dielectric values to Hounsfield units, which are present in computed tomography (CT) scans, and to MRI intensity values, which are present in magnetic resonance imaging (MRI) scans. The code successfully worked for single frequencies and over a range of frequency values (using a Debye model). This allows for doctors to understand images producing using electromagnetics in readily understood CT and/or MRI formats without requiring any additional training, leading to timely and accurate medical diagnosis. The conversion method developed allows for existing medical diagnostic tools and analysis techniques to be used directly with microwave imaging. In addition, the method for conversion from an image in Hounsfield units to dielectric values and conversion from an image in dielectric values to Hounsfield units can go in both directions. Furthermore, the method for conversion from an image in dielectric values to MRI intensity values includes creating a water content map and a T1 map as an intermediary step. The patent also included a method to convert medical images in Hounsfield units to dielectric values using a frequency dependent model. Deriving dielectric models from CT scans is often useful when solving complex problems in computational electromagnetics.

The second patent titled “Distributed Microwave Image Processing System and Method” resulted from the need to want all imaging centers, radiology groups, and/or doctor’s offices to be able to have access to images produced using electromagnetics without having to upgrade their computer hardware. A method was developed to allow for the majority of image processing and image reconstruction of microwave images to occur in a centralized computing environment. Instead of performing image processing and image reconstruction at the imaging centers, radiology groups, and/or doctor’s offices, these remote sites send the microwave data they collect to the the centralized computing environment.  The centralized computing environment also offers another distinct advantage; the data and results acquired at all the remote sites can be stored and used to enhance processing and reconstruction of microwave images. The centralized computing environment takes advantage of multiple processors to perform iterative reconstruction and seeds the reconstruction using prior data. In one embodiment of the invention, the seed is generated by first comparing collected and stored scattering fields to find a best or closest match and then using stored data of a prior reconstructed image reconstructed corresponding to the stored scattering fields of the best or closest match. In another embodiment of the invention, the seed is generated by both of (1) using the collected microwave data and (2) using stored data of a prior reconstructed image of a different patient which closely matches data of the current patient. The centralized computing environment also has the capability to convert medical images in dielectric values to Hounsfield units. The method developed and described allows for more accurate image reconstructions to occur in less time than if they were performed at remote sites.

It is exciting to work on new technology and methods that can have a real impact on the health of patients. Below are three patent certificates that were created to celebrate the accomplishment of having these three patents granted.

The post Description of Three Patents Named Co-Inventor On Assigned to Ellumen Inc appeared first on Todd McCollough's Website.