collecting data/developing/clinical trials/licensure), we have the experience and expertise of ... most up to date netwo
COMPANY PROFILE
We believe that AI can be of great value to doctors. VUNO products will not only provide improved workflow but a whole new user experience.
2017
Contents 01 About VUNO 02 Our Technology 03 Key Offerings
VUNO |
| VUNO
About VUNO VUNO We are quickly approaching an era where AI has taken over the world. The medical field is certainly not an exception, with its various problems being solved and improved by the help of AI. Based on the vast majority of data collected, some AI-based models perform and imitate similarly as, if not better than, humans. However, we believe that the core value that AI brings to the table is not just about its accuracy, that far exceeds human abilities. Through AI, we can provide tools that are ready to be used at any moment to improve your workflow—with tools that it has never seen before. Acknowledging that AI has reached almost every aspect of our life, we pursue affordable products that you can comfortably choose from. After all, it is the true needs of doctors that we take into careful consideration when we create solutions for problems that AI can take on.
COMPANY PROFILE
“ We believe that AI can be of great value to doctors. VUNO products will not only provide improved workflow but a whole new user experience.”
4
COMPANY PROFILE
5
VUNO | About VUNO
About VUNO | VUNO
VUNO-MED™
Our Expertise
The VUNO-MED™ series will foster a work environment for medical staff that will help expedite the decision-making process with better accuracy. Based on data collected from a number of videos and signals from various medical fields, we have come up with ways to provide diagnostic assistant services. The VUNO-MED™ series is a collective collaboration of doctors and engineers with years of experience in medical data analysis, as well as those with various medical experiences. From X-ray features and CT scans to a biosignal monitoring system, the series provides a variety of services such as quantifying and assisting with diagnosis of diseases, as well as analyzing data. With the VUNO-MED™ series powered by VUNO, you can experience the future of medicine.
Expert experience and knowledge in medical innovations · We have radiologists and medical scientists as advisory members on board with us, with an in-depth collaboration and cooperation of data analysis specialists and UI/UX designers working on our products. · Established in 2014, we have reached more than 10 major hospitals in Korea, as we continue to develop ideas on numerous possibilities made through the process of codevelopment, clinical trials, and verification. We have the experience and a philosophy on how technology can provide the needs of hospitals and doctor. · We have issued 14 patent applications on the topic of innovative technology and presented more than 10 international clinical trials with Service Corporation International (SCI). Currently, we are in the process of getting a variety of medical devices with AI-based diagnostic software (CADx/CADe) licensed. · We have participated in establishing guidelines for licensing AI-based medical devices. Currently, we are in the process of implementing advice from the Ministry of Science and Technology, Food and Drug Administration, and Federal Emergency Management Agency.
Independent technology to support the entire cycle of analyzing various forms of medical data · We have secured our very own deep-learning engine (VUNO NET) as well as our own data collecting and analyzing platform (VUNO-MED™). By using these, we have analyzed various data and diseases and invented different models through learning. · From data collection, preprocessing, reevaluation/relabeling to optimization of the deep learning network, we have the software and infrastructure to support the full process, which has been validated through multiple clinical cases.
Commercialization and development of products that are ready to launch in the market · Having completed the entire pre-process prerequisite for doctors to use in real time (strategizing/ collecting data/developing/clinical trials/licensure), we have the experience and expertise of not only conducting but managing on our own. We bring AI-based products to life as a real, legitimate medical equipment for our customers. · Our specialization lies in proactive, transparent communication with our customers and clients to best understand and solve any potential problems that they may experience. With the VUNOMED series, you get to decide how to pursue the results that you want.
COMPANY PROFILE
6
COMPANY PROFILE
7
VUNO | About VUNO
About VUNO | VUNO
History
Partners
Since the foundation of VUNO in December 2014, We are moving ahead to innovate artificial intelligence.
2014
Establishment of VUNO Seed investment and inducement
( USD 1million, Bonangels/futureplay/TIPS)
2015 Establishment of Research Institute (Korea Industrial Technology Association)
Completion of VUNO NET the Deep-learning engine
2016
Selected as one of the top 5 for ImageNet ILSVRC 2015 Classification Task
Development contract on AI-based diagnostic devices, signed and sealed with 8 hospital in major cities Series A investment and inducement ( USD 3million, SBI Investment, Smilegate Investment, HB Investment )
KGMP Certified
(Ministry of Food and Drug Safety; response to ISO13485)
Backers
Participation and presentation at Nvidia GTC/RSNA Exhibitions
2017 Development contract on AI-based diagnostic devices, signed and sealed with 4 regional hospitals in major cities Presentation of academic journal and thesis at RSNA, C-MIMI, AJR, JDI, etc. Completion of VUNO-MED™ Bone Age development and in process of clinical trials (Ministry of Food and Drug Safety)
Participation and presentation at NvidiaGTC/SIIM/RSNA Exhibitions
COMPANY PROFILE
8
COMPANY PROFILE
9
VUNO |
| VUNO
Our Technology Deep Learning Engine VUNO NET Our engine exclusively for analyzing medical data Although deep-learning models show impressive performances, there are many things to consider when commercializing products. For example, a number of algorithms require extensive computing resources which can be problematic for a commercialized product. Invented in 2015, VUNO NET the deep-learning engine based on CNN/LSTM, is one of many that we have developed utilizing the most up to date network architecture and processing technique. VUNO NET is compatible with any software/hardware environment, and we can guarantee the fastest in the lightest form. In addition, VUNO NET has been evaluated over many years by over a 100 different architectures on GPU machines like Titan X or P100.
COMPANY PROFILE
10
COMPANY PROFILE
11
VUNO | Our technology
Our technology | VUNO
Platform for medical data collection and model development, VUNO-MED™
Optimized algorithms made from experience and deep understanding in medical data
Based on medical data, the initial process of data collection with model development and its optimization is crucial to solve any given problem that may surface. Built on cloud-based data, VUNO is an exclusive platform that supports optimization through data collection/relabeling and model learning. We provide the best environment for the application of deep learning in medicine, where many of the collected image data may require reviewing and relabeling from a experts. We understand how costly the entire process of monitoring and reevaluating data can be. Therefore, we would like to provide you the best environment to make this process easier and more efficient. Compatible with Linux, Windows, and OS, VUNO-MED™ is the final product and piece of invaluable experiments, trials, and reviews conducted by various expert organizations and officials.
We have profound understanding of deep-learning and machine-learning technology. We have also designed an independent architecture that can produce a top-notch performance against any given problem. In 2016, we were selected as one of the top 5 competitors at the world’s biggest, ImageNet’s own ImageNet Large Scale Visual Recognition Competition (ILSVRC). Through major news networks, we were introduced as the deep-learning company worth noting. After collaborating with more than 10 major hospitals and collecting more than millions of visual data from X-ray/CT/MRI scans, we have developed various deep-learning models to analyze, classify, detect, and quantify data, which has led to commercialization of a variety of diagnostic software devices. The following explains what separates our products from others from a technological viewpoint:
Our platform includes data acquisition and labeling tools
1
Data Collection & Labeling Modules
2
Medical Image Processing Modules
Web-based Labeling Tool
API Server
3
Deep Learning Framework
WebGL
cuDNN
C++ MKL BLAS
Labeled Image
GPUs
ATLAS
CUDA Open BLAS
CPU
By utilizing our VUNO NET engine, we can embed any shape of medical device, including computing conditions with insufficient resources, and provide services where any device can thrive.
Deep Learning Applications
VUNO NET
Database
CPU
The strongest yet lightest module
cuBLAS
cuFFT
Minimization of variance between institutions/image settings Algorithms developed based on a deep understanding of each X-ray/CT machine’s unique visual characteristic allows us to provide a universal performance for images from any equipment in any institution.
Deep Learning Primitive Library
Language & Math Library
Learning algorithms for various labeling
In the process of data labeling with a large number of experts, we inevitably run into a difference in opinions, which then usually leads to following the majority. However, we see the difference of opinions as another way to produce meaningful information from doctors with different, diverse experiences of labeling.
nVIDIA GPUs
Operating System (Linux/Windows)
Clinically-validated, medical image-specialized module We provide modules for quantification/detection/classification/search essential for medical image analysis, as well as commercialization-qualified services for algorithms and specific diseases. Each module has gone through both verified clinical trials and evaluation. Clinically oriented UI/UX We produce optimal UI/UX designs that considers the actual clinical working environment and the hospital’s system. In the end, we believe that even the best technology is useless without a good user interface.
‘Future Human AI’ MBC
COMPANY PROFILE
12
COMPANY PROFILE
13
Our technology | VUNO
VUNO |
Scientific Papers Based on deep-learning models, VUNO has developed products derived from various medical image data as well as medical image analysis and diagnosis. The innovative results that we have suggested have been presented in the highest academic journals and institutions in different fields. Our AI-implemented products can be of huge help to delivering accurate and efficient diagnosis as well as improving the clinical workflow.
· Accuracy Enhancement with Deep Convolutional Neural Networks for Classifying Regional Texture Patterns of Diffuse Lung Disease in HRCT / RSNA 2016
· An Automatic Classification Platform for Differentiation of Regional Diseased Patterns of Diffuse Infiltrative Diseases
on High Resolution CT Using Lung Segmentation, Support Vector Machine and Convolutional Neural Net Classifications / RSNA 2016
· Automated Opportunistic Screening for Osteoporosis Using Abdominal CT Exams and Deep Learning Artificial Intelligence / RSNA 2016 · Deep learning-based content-based image retrieval for finding HRCT images of similar patients with interstitial
lung disease: Validation with 100 paired HRCTs and automatic quantification of six disease patterns / RSNA 2017
· A Comparative Study of Automatic Hand Bone Age Assessment Systems / RSNA 2017 · Computer-assisted program using deep learning technique in determination of Bone Age:
evaluation of the accuracy and efficiency / American Journal of Roentgenology(AJR) 2017
· False positive reduction by actively mining negative samples for pulmonary nodule detection in chest radiographs / C-MIMI 2017 · Comparison of shallow and deep learning methods on classifying
the regional pattern of diffuse lung disease / Journal of Digital Imaging(JDI) 2017
· Retinal vessel segmentation in fundoscopic images with generative adversarial networks / arXiv preprint 2017 · Towards Accurate Segmentation of Retinal Vessels and Optic Disc in Fundoscopic Images with
Generative Adversarial Network, Investigative Ophthalmology & Visual Science(IOVS) / Under Revision 2017
COMPANY PROFILE
14
COMPANY PROFILE
15
VUNO |
| VUNO
Key Offerings Offering Map
< VUNO-MED™ >
We have been brainstorming how to directly apply AI in the medical field. In particular, we have been designing products for X-ray/CT images that will allow doctors to make decisions effectively using the information provided by our AI products. Our products can be easily adopted and utilized in complex medical environments. We present our commercialized products to you with much confidence.
CT
X-ray
BONE AGE
CHEST SCREENING
LUNG QUANT
VUNO-IO (Labeling Platform)
COMPANY PROFILE
16
COMPANY PROFILE
17
VUNO | Key offerings
Key offerings | VUNO
VUNO-MED™ Bone Age A Bone Age study is the process of assessing and detecting the degree of maturation of a child’s bones and growth. Based on the morphological characters and stage of the bone formation, the main purpose is to obtain one’s Bone Age and analyze the difference with the actual age of the bone. There are various evaluation methods such as G.P. (Greulich and Pyle) Atlas, TW (Tanner-Whitehouse) and localized transformation. The G.P. Atlas-based method of determining the age according to one’s sex (male and female) is the most popular method around the world. However, Bone Age examinations are very subjective and one’s experiences can affect the results as well, regardless of the method. In particular, the process of finding and comparing images from references can be a very routine and arduous task for doctors, especially at large hospitals. It can be an even more difficult task for doctors who don’t do this on a regular basis. To solve this problem, we have created our very own VUNO-MED™ Bone Age products to provide and guarantee better accuracy for doctors with little experience and to expedite the process for those who perform this task on a regular basis.
‘ChosunBiz’
“
We used to spend few minutes per case, but with the VUNO-MED™ Bone Age products, we were able to reduce this to 5 seconds per case, which allowed us to quicken the whole process - Dr. Jin-sung Lee at the Department of Pediatrics of the Asan Medical Center, Korea
”
VUNO-MED™ Bone Age · We have verified that the use of VUNO-MED™ Bone Age has decreased the time spent on reading scans by 40% for doctors with little experience and by 20% for experienced doctors.¹ · G.P. (Greulich and Pyle) Atlas-studied and listed by SW-determined age in order of probability, the area in which the judgment is based on is visually shown. · MAD (Mean Absolute Difference): 6.47 Month ² · Supported formats DICOM/PNG/JPG
Comparison of Reading Time for Bone Age Assessment Reviewer A (Experienced fellow) Manual VUNO-MED™ BONE AGE Assisted
Reviewer B (Second-year resident)
Reading time(min) 0
50
100
150
200
1) Computer-assisted program using deep learning technique in determination of Bone Age: evaluation of the accuracy and efficiency, American Journal of Roentgenology(AJR), 2017 2) The difference in average between results determined by SW-decoding and the reference standard reviewed by 3 different decoders, which essentially means the difference of results between human decoding and SW-decoding COMPANY PROFILE
18
COMPANY PROFILE
19
VUNO | Key offerings
Key offerings | VUNO
VUNO-MED™ Chest X-ray
The necessity of VUNO-MED™ Chest X-ray Often used for screening purposes, Chest X-ray screening is a very frequent medical practice. However, the probability of human error is high due to the limited amount of information provided by modality, as well as the difficulty in reading. -> AI-based Screening S/W can be used to improve the accuracy, as well as the efficiency of radiological reporting.
VUNO-MED™ Chest X-ray is a chest-disease screening system that assists diagnosis by providing screening results and suspected radiological findings from chest radiographs. With Multi-task learning, you can classify chest-related lesions and suggest the location of these lesions to maximize the efficiency of radiological reporting. With a UI that produces an automatic report based on the AI’s initial findings, our product provides a whole new user experience.
Automatic location detection for six lung pathologies · Objects Nodule, Consolidation, Effusion, Opacity, Pneumothorax, Emphysema · Normal / abnormal screening based on the 6 pathologies · Visualization of suspected lesions · Automatic generation and selection / correction of reports for detected pathologies Providing the revolutionary AI-enabled UX We have closely observed the reporting process and the hospital’s workflow. Once AI analyzes the images and delivers information of diagnosis-related references to the doctor, our UX allows the doctor to generate a final report quickly with minimum effort.
‘Future Human AI’ MBC
COMPANY PROFILE
20
COMPANY PROFILE
21
VUNO | Key offerings
Key offerings | VUNO
VUNO-MED™ Chest CT
Next-generation lung nodule detection performance · High sensitivity with low false positive rate
VUNO-MED™ Chest CT is a detection solution for pulmonary nodules in CT thorax scans. Finding a nodule from an LDCT scan is very common and it often requires more time than x-ray reporting. VUNO-MED™ Chest CT is an optimized product for faster, more accurate reading of such cases. The nodule detection technology through CT scans has existed before, but was not able to expand significantly due to its limited performance. AI-technology breaks down these technological barriers through learning of vast historical data. Experience the unprecedented high accuracy of detection with our VUNO-MED™ Chest CT.
· Adjustable operating points for various settings and environments
Pulmonary nodule detection using hybrid-multi contextual deep learning approach From raw CT scans, Identify regions of possible nodule and provide a nodularity Features · High sensitivity with controlled false positive rate · Detection of actionable nodules from 3mm to 40mm in size · Suggestion of approximated size of nodule · Customizable operating setting between sensitivity oriented environment for high risk patients and specificity oriented environment for efficient screening Performance Validations · Well-known public datasets : LIDC/LUNA16, Kaggle DSB Lung Cancer Dataset(labeled by 3experts) · Sensitivity analysis of high risk patients with Asan Medical Center(The largest hospital in Korea) · Screening analysis of low risk patients with Kangbuk Samsung Hospital (The largest medical checkup center in Korea)
Research and Development Timeline
1st clinical validation (10,000+ cases)
2nd multi-center clinical validation (150,000+ cases)
KFDA/FDA/CE Thoracic Radiologist
Initial model development
3Q 17’
‘Future Human AI’ MBC
COMPANY PROFILE
22
COMPANY PROFILE
General Radiologist
4Q 17’
1Q 18’
2Q 18’
3Q 18’ 23
VUNO | Key offerings
Key offerings | VUNO
VUNO-MED™ Lung Quant
The six disease patterns The method of treatment of DILD can vary according to the six disease patterns, so it is important to classify the pattern correctly. -> Normal, Ground-Glass Opacity(GGO), Consolidation, Reticular Opacity, Emphysema, Honeycombing
VUNO-MED™-Lung Quant is an automatic diagnosis solution to provide accurately classified and quantified information, analyzed and derived from each lesion pattern from a patient with DILD (Diffuse Interstitial Lung Disease) and his or her HRCT (High-Resolution CT) visuals. We can diagnose lesions with high accuracy based on learning from high quality labels created by some of the best specialists in the field. VUNO-MED™Lung Quant is a secondary diagnostic solution that retrieves similar cases from a pre-analyzed database and provides multidimensional information through quantified information of analyzed lesions.
(a) Normal lung parenchyma (b) Ground-glass opacity (c) Consolidation (d) Reticular opacity (e) Emphysema (f) Honeycombing
[Disease pattern classification accuracy]
[Whole lung quantification agreement]
97% 85%
Conventional clinicallydefined features
82% 70%
Our datadriven features
Between two Between a radiologist experienced radiologists and our result
· Dataset : Asan Medical Center based in Seoul, Korea, Siemens and the National Jewish Healthcare Center based in Colorado, US, GE. · Reference : Comparison of shallow and deep learning methods on classifying the regional pattern of diffuse lung disease / Journal of Digital Imaging(JDI) 2017
COMPANY PROFILE
24
COMPANY PROFILE
25
VUNO | Key offerings
Key offerings | VUNO
Key Takeaways
Better performance with CAD(Computer Aided Diagnosis) and AI enabled UI/UX
COMPANY COMPANY PROFILE PROFILE
AI Platforms for Data-driven solutions
Improved clinical workflow by eliminating routines
26
COMPANY PROFILE
Reliable solution with Multi-clinic validation
27
VUNO |
| VUNO
Visit VUNO at RSNA 2017 Booth #8510 North Hall
COMPANY PROFILE
Booth #8149R ML Pavilion
28
COMPANY PROFILE
29
VUNO www.vuno.co |
[email protected] | +82-2-515-6646 6F, 507, Gangnam-daero, Seocho-gu, Seoul, Korea
