Imaging Phenomics for Precision Medicine MichaÍl AUFFRET Chief Product Officer – Median Technologies Michael.Auffret@mediantechnologies.com
Transforming imaging data into actionable clinical information Our Mission: Enable personalized and predictive medicine through the development and marketing of innovative medical image analysis solutions based on artificial intelligence and cloud computing
Our Beginning: Pioneering the industry – extracting the most meaning out of medical images since 2002 Our Business: Best-in-class technology, including image processing technology, cloud computing & artificial intelligence, leverages the power of Imaging Phenomics to improve current treatments, enhance diagnostics and accelerate the development of next generation therapies Our Growth: Powered by proprietary technology, robust Key Opinion Leader (KOLs) connections and strong medical and scientific partnerships and collaborations Our People: 100 employees worldwide across Europe, US and Asia; headquarters in Sophia Antipolis, France and offices in Woburn, MA (US), Shanghai and Hong Kong, China 2
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Precision Medicine • Goals of better predicting disease risk, understanding how diseases occur, and finding improved diagnosis and treatment strategies • Provide the best available care for each individual and refers to the stratification of patients into subsets with a common biological basis of disease • The industry needs a more comprehensive approach to personalized medicine that uses phenomics, genomics, and big data to identify the best treatment for each patient 3
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Source: Lambin et al , Eur J Cancer. 2012 Mar;48(4):441-6.
Unmet Medical Need • Huge volumes of medical imaging data are generated each year (~2 trillion images/year globally, 2 billions + of imaging exams)
• Clinicians need tools and clinical data registries for precision medicine • Biopharma companies need large volumes of imaging data to analyze patient response & identify companion diagnostics • Clinical and imaging data is siloed in individual healthcare institutions, unstructured, and un-curated, preventing clinicians from utilizing data from previous patients to inform treatment decisions • In the near future all imaging exams will be post processed in the cloud using AI. Growth in the AI health market is expected to reach $6.6 billion by 2021 (Accenture) 4
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Median proprietary technology and platforms AI helps automate a number of labor intensive tasks, reduce human errors and ensure data accuracy Configurable technology platform with computer-assisted lesion detection (CAD), identification and follow up of lesions in medical images
5
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
AI impacts treatment decisions
Next generation high-throughput phenotyping platform to acquire, index, and analyze thousands of individual phenotypes from medical images and clinical data utilizing unsupervised machine learning
From Genotype to Endotype
6
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Imaging Phenomics
Decoding the tumor phenotype using non-invasive imaging • Scientific evidence supports the hypothesis that genetic and/or molecular alterations within tumors manifest as specific, macroscopic, observable changes in imaging signatures • Imaging phenomics is the systematic, large scale extraction of imaging features for the characterization and classification of tissue and disease phenotypes.
HCC
Vessel
• Imaging phenomics can prospectively help identify patient subtypes that may benefit from specific therapies • Data driven imaging phenomics are expected to play increasingly important roles in the development and clinical assessment of targeted therapies 7
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Artifact
ITK/VTK Image Viewer
iBiopsy® Platform
Proprietary & patented cloud-based AI technology & processes Extraction, Indexing, Search and Data Analytics Platform Component Phenotype Signature and Profile Matching
Phenotype image database
Imaging Features extraction & clustering
Phenotype signature database
Automated Indexing (offline) iBiopsy® totally unsupervised, automatic and asynchronous extractions of features from the images and indexing them in a no-SQL database based on unique similarity metric.
8
User Request
Most similar phenotypes from reference database are returned to the user
Real-Time Search (online) CyberScan® performs real-time similarity searches of millions of indexed phenotypes against a target patient phenotype
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
The Sherlock™ analytics engine is the inferencing engine which provides advanced data analytics, deep learning and advanced clustering to support the detection of abnormal pathology, lesion classification and disease prediction
Data Collection and Indexing The iBiopsyÂŽ Platform will aggregate, curate, search, and analyze medical image data Images + Clinical Data
Healthcare Partners
Phenotype extraction Unsupervised, automatic and asynchronous extractions of features from images and indexing in a no-SQL database based on unique similarity metric
Liver
3
Prostate
Data Curation Lung
BioPharma Partners Other Indications
9
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Imaging Features extraction & clustering
Storage
Clinical Data Registry for HCC • Acquire, process, curate and index large amounts of clinical data obtained from major clinical centers as part of routine clinical practice.
During late arterial phase on CT, HCC appears brighter than surrounding liver
• Improve diagnosis and prognosis of HCC and early detection of HCC. • Take into account the underlying molecular basis of the disease.
In portal venous phase, HCC appears darker than surrounding liver (washout)
• Predict prognosis and guide therapy is an essential development in clinical oncology to improve outcomes. clinicaloptions.com
10
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
The Approach: Dimensionality Reduction Compact Image-Based Signatures Automated organ segmentation in ‘tiles’ with a unique signature computed to each tile.
A tissue classification map is computed for the image, based on similarity between the signatures.
Profile matching in real time
11
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Image Signature Development Roadmap A stepwise approach
Phase 1
Procedure
Endpoint
12
Phase 2
Imaging Cluster Discovery
Imaging Cluster Robustness
The number of lesions observed with imaging phenotype maps based on CT scans concordant with the histopathological confirmed lesion number (lesion size > 10mm).
• Reproducibility of the imaging signatures. • Repeatability of the image signatures (test retest on the sequential scans performed on the same patients).
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Imaging Cluster Biological Association
The number of imagebased phenotypes concordant with the underlying pathophysiological processes in liver lesion
Imaging Cluster Effectiveness
The number of robust image-based phenotypes associated with clinical outcome for large healthcare deployment
Objectives iBiopsyÂŽ Phenotyping Platform Development Building Validated Clinical Data Registries (CDRs) to assist clinicians through image-phenotype signatures Search System in characterization of liver lesions
Validated CDRs The method consists of two phases:
1. An offline CDR construction phase. 2. An online real-time query phase with summary statistics based on retrieved results.
13
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
HCC Project - Validation Technical Validation in Process • Acquisition Parameters: Datasets acquired with different slide thicknesses and different iterative reconstructions (3 and 4). • Repeatability: Patients with various lesions that had several scanners in less than a month. • Reconstruction: Datasets reconstructed with different levels of iteration (ID2, ID3, ID4) with same slice thickness (2 mm) • Datasets reconstructed with different slice thicknesses (2mm every 1, 3mm every 2 then 5mm every 3) • CTs of HCC operated with histopathology with evaluation of fibrosis, activity and steatosis • CT with metastases • CT of HCC on healthy liver
14
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France
Our Core Values Leading innovation with purpose Combine the spirit of innovation with our passion and conviction to help cure cancer and other debilitating diseases.
Thank you!
Committing to quality in all we do Be dedicated to quality in everything we do. Quality begins with us and we are committed to it.
Supporting our customers in achieving their goals Listen to the needs of our customers and help make their goals our goals through our innovation, imaging expertise, superior services and quality solutions.
Putting the patient first There is a person at the other end of the images we analyze who is counting on us to do everything we can to help make them healthier.
15
www.mediantechnologies.com | SophI-A Summit | Nov. 7-9, 2018 | Sophia-Antipolis, France