Short Article

AI and Precision Medicine: Improving the Patient Experience

Deep learning and medical imaging advances are combining to deliver significant improvements in diagnostics and precision medicine.1 With predictions of an $8 billion market by 2027, there is tremendous potential for precision medical imaging to transform and evolve into care pathways that are more effective, personalized, predictive, and targeted.1

One of the biggest challenges in healthcare big data analytics has been extracting and analyzing data from static images and organizations are looking to develop artificial intelligence (AI) solutions to make better use of image data.2 AI-enriched imaging equipment will help imaging protocols and procedures adapt and become more personalized.1 There is so much additional data contained in CT, MRI, and other modalities that can be often unusable without deep learning, however, recent advances in pattern recognition mean subtleties that may be missed by the human eye are now identifiable by machine learning methods.2

With many collaborations and partnerships in the works to address challenges inherent in precision medicine achievements, here is a look at a few of the intriguing projects that are impacting the intersection of precision medicine and medical imaging.

Non-pediatric specialists can make accurate diagnoses sooner

With recent developments in AI and subsequent deep learning algorithms, precision medicine, a preferred method for treatment planning, can now enable healthcare practitioners, like radiologists, to make informed decisions based on millions of previous patient images that meet a similar profile.

In a collaboration between medical imaging manufacturers and radiologists from a children’s medical facility, a new cloud-based decision support platform hosting millions of reference scans will help in the identification of the large variability in pediatric brain MRI scans.2 The goal is to increase diagnostic accuracy and provide real-time contextual information for physicians to differentiate between concerning conditions and natural changes at different stages of a child’s brain development.2

For example, changes in myelination that takes place during the first few years of life is often confused with disease states or can lead to misinterpretation of exam results as normal when the abnormality is symmetric in the brain.2

Cloud technology will be used to process high-volume data and images while radiologists will help develop the new platform, which will provide physicians with reference scans of children of all ages to assist in interpretation.2 Most pediatric brain imaging is not performed by specialists in children’s hospitals where an expertise in understanding the brain development is critical for accurate interpretation.2 Providing non-pediatric specialists access to the necessary knowledge and expertise via the new platform at the time of interpretation will boost confidence and performance as well as the likelihood of more accurate diagnoses.2 Additionally, children are less likely to be sent on for further testing and life-saving treatment can begin sooner.2

Development of this brain app is part of a library of deep learning algorithms intended to help clinicians make quicker and more accurate diagnoses for common and complex medical conditions using imaging.2

Clinical decision support for oncology

One major obstacle to wider adoption of precision medicine has been access to integrated large volume datasets, many of which exist but are stored separately in private databases.3 Development of deep learning algorithms necessary to move individualized diagnosis and treatment planning forward also require vast amounts of training and testing data.3 Although this data exists, it needs to be incorporated into tools that provide informed decision-making for healthcare practitioners needing to select the right treatment option for the right patient in real-time.3

To make progress in addressing data sharing challenges, an industry collaboration combining the expertise of medical imaging and diagnostics experts is creating a platform that integrates data for precision health in oncology.3 As a first of its kind oncology platform, it combines three aspects: radiological imaging, “in-vivo” data acquired directly from the patient to characterize the tumor at the anatomical and physiological level through monitoring equipment and “in-vitro” information from laboratory tests that characterize the tumor at the molecular level based on tissue pathology, blood-based biomarkers, genomic alterations (cancer-relevant mutations) and other factors.3 Electronic medical records, medical best practices, and information on the latest research are added for broader context.3

Currently, the process for treating oncology patients is an inefficient one where a team of specialists meet at a scheduled time to share and discuss details, including test results, patient status, diagnosis, and disease progression assessments via spreadsheets, slides, and images.3 By contrast, a comprehensive data dashboard on the new platform will pull together diagnostic imaging, gene sequencing, tissue samples, blood tests, and the latest clinical trials, along with other relevant data that supports decision making in one place.3 When physicians have this level of information at their fingertips they are able to more accurately and quickly determine a diagnosis as well as the most effective next step in the treatment pathway for a particular patient.3

Predictive analytics and AI apps in the platform compare a patient’s data with historical treatment data and outcomes and are accessible by the care team wherever they are.3 To take this concept even one step further, the patient’s longitudinal data from seven years through the present moment is incorporated so oncologists can make appropriate treatment choices through every disease stage.3

Another benefit of leveraging big data analytics to predict next steps with precision medicine means physicians have an abundance of fresh insights to help them adapt gold standard care to individual patients because they have the information necessary to know if the patient’s genetic makeup is likely to respond to a given treatment.3

Advantages expected from using machine and deep learning to leverage and combine massive datasets include reducing unnecessary biopsies and other procedures due to suspicious findings.3

Embedding adaptive intelligence tailors imaging to patient

Imaging analytics is a growing area of interest for health IT vendors and developers interested in decoding the massive amount of data contained in medical imaging modalities such as CT, MRI, and ultrasound, among others.4 One reason may be that recent advances in deep learning pattern recognition algorithms have started to make it easier for computers to understand what images represent and how to identify subtle variations in anatomy that can escape the expert eye of human clinicians.4

One partnership that’s focused on the future of healthcare where technology, analytics, and cloud computing power combine to help physicians make faster decisions leveraging the technical expertise of a global medical technology provider and the clinical and research expertise of one of the largest recipients of National Institutes of Health (NIH) funding.Together they will apply deep learning strategies to smart imaging analytics and clinical decision support apps in order to make the promise of precision healthcare a reality for patients with complex needs.4

The initial focus of the partnership is a project to develop a library of deep learning algorithms that are accessible through a cloud technology partnership and designed to connect smart imaging machines in order to identify pneumothorax, or collapsed lung, in trauma patients as quickly as possible.4 The project uses machine learning to fine-tune the process of distinguishing between normal and abnormal scans, providing clinicians with sensitive and accurate information in near real-time about patients at risk for serious complications from the life-threatening condition and recommended actions.4 Predictive analytics, precision medicine, and increased automation of routine care delivery are planned for subsequent projects.4

Combining analytics, deep learning and cloud technology is especially powerful because they become smarter and more scalable over time.4 Eventually, electronic health record data will be integrated into deep learning algorithms to enhance the sensitivity of the algorithms and analytics tools.4

Looking forward

Combined approaches the examples above enable better, more informed decisions to be made by clinicians and researchers. And while the possibilities with AI and precision medicine seem endless, there remain challenges on the people side -- training clinicians and optimizing this data for clinical use remains a challenge5 -- all signs indicate that AI and precision medicine will be game-changers across nearly every area of medicine. What was once the wave of the future is already here.

REFERENCES:

  1. Medical Imaging, Machine Learning to Align in 10 Key Areas. Health IT Analytics https://healthitanalytics.com/news/medical-imaging-machine-learning-to-align-in-10-key-areasAccessed 5/24/2019
  2. [USE CASE] Boston Children’s, GE Develop on Smart Imaging Technology. HIT Infrastructure https://hitinfrastructure.com/news/boston-childrens-ge-develop-on-smart-imaging-technologyAccessed 5/20/2019
  3. [USE CASE] GE, Roche Partner for Big Data Analytics, Precision Medicine Platform. Health IT Analytics https://healthitanalytics.com/news/ge-roche-partner-for-big-data-analytics-precision-medicine-platformAND https://www.ge.com/reports/digital-medicine-ge-roche-will-analyze-medical-data-find-better-treatment/ Accessed 5/20/2019
  4. [USE CASE] UCSF to Develop Machine Learning for CDS, Imaging Analytics. Health IT Analytics https://healthitanalytics.com/news/ucsf-to-develop-machine-learning-for-cds-imaging-analyticsAccessed 5/20/2019
  5. How AI Systems Can Improve Medical Outcomes. Knowledge @ Whaton. https://knowledge.wharton.upenn.edu/article/ai-based-systems-can-improve-medical-outcomes/. Accessed 6/18.