■ Line-up of Konica Minolta Products for primary care ■ Create movie-like images with use of pulse x-ray. ■ X-ray condition (ex.)[r]
(1)U d f K i Mi l ’ AI Cli i l D i i The 21st Annual Vietnamese Congress of Radiology and Nuclear Medicine, 2019
Satoshi KasAI
Update of Konica Minolta’s AI Clinical Decision Support Software
Satoshi KasAI
Clinical R&D Division, Image Processing Technology Department X-ray Business Operations,
Healthcare Business Headquarters
History of AI
1960s 1970s 1990s 2010s
～ Winter Era ～ Winter Era ～
1980s 2000s
Limitations on teaching expert knowledge Limitations on solving
practical problems Search and Inference
- The word “Artificial Intelligence” was created
- Basic neural network was introduced
Knowledge-based AI with expert system
- Productization in a part of industrial area Machine learning based AI- Deep Learning
★”Automated diagnosis” was discussed in a paper
★Concept of computer-aided diagnosis (CAD) was created
1995 SIANN （U of Chicago)
1998 R2 Technology 1stCAD product
KM CAD products
- CAD for mammogram（2010） - Temporal Subtraction（2013） - Bone Suppression（2015） ■Neural Network
Launched in 2010 to Japanese market
(2)Target Diseases of Medical Image
Mammogram Chest X-ray
Breast cancer Lung Cancer, (Breast Cancer) TB TB Pneumonia Cardiomegaly Pneumothorax PE Atelectasis Rib fracture Rib fracture Effusion Emphysema Fibrosis Heart failure Hernia
AI technology in 2nd Generation
a ge re-essi ng tur e ct io n s if ica o
n sult
Input Output
✔Conventional method
Numerical conversion
A l O
Im
a Pr
pr oc e Fea t ext ra Cla s s ti o Re s Ex.) region extraction
Color red orange Size large small Shape circle ellipse TextureSparse white dot Dense dots
RG B Area (pixel number) Circularity SD Apple Orange (158,46,36) (242,118,26) 100cm2 (2000) 25cm2 (500) 0.90 0.86 11.1 20.2
Original Image Bone
Suppression
Bone Suppression
© 2016 Konica Minolta, Inc
Bone Suppression
(3)History of AI
1960s 1970s 1990s 2010s
～ Winter Era ～ Winter Era ～
1980s 2000s
Limitations on teaching expert knowledge Limitations on solving
practical problems
Search and Inference - The word “Artificial Intelligence” was
created
- Basic neural network was introduced
Knowledge-based AI with expert system
- Productization in a part of industrial area Machine learning based AI- Deep Learning
★”automated diagnosis” was discussed in a paper
★concept of computer-aided diagnosis (CAD) was created
1995 SIANN （U of Chicago)
1998 R2 Technology 1stCAD product
KM CAD products
- CAD for mammogram（2010） - Temporal Subtraction（2013） - Bone Suppression（2015） ■Neural Network
Conventional vs Deep Learning
a ge re-essi ng tur e ct io n s if ica o
n sult
Input Output
✔Conventional method
Im
a Pr
pr oc e Fea t ext ra Cla s s ti o Re s
e sing re ion fica ult
(4)Target Diseases of Medical Image
Mammogram Chest X-ray
Breast cancer Lung Cancer, (Breast Cancer) TB
TB Pneumonia Cardiomegaly Pneumothorax
PE Atelectasis Rib fracture Rib fracture Effusion Emphysema
Fibrosis Heart failure
Hernia
■Will show Chest AI Demo
① Several usage of chest AI : Triage #1
- AI will prioritize cases Cases with high priority will be
Notification
AI
Radiologist
- Cases with high priority will be delivered and noticed to doctors right after image creation
✔Value hypothesis
- Reducing time for physician to access abnormal cases - Next action could be determined within a same day
- The image will be interpreted by a radiologist - Simultaneously, AI checks
② Several usage of chest AI : Q/A
Diagno sis Same
the image
- In case the results of doctors and AI are not matched, those cases will be sent to 2nddoctors (or
same doctor)
Radiologist
AI
Radiologist or NLP
2ndor same Radiologist
sis
Discrepancy
Check
16 ✔Value hypothesis
(5)- AI will be used as 1streader
- In case images are classified as normal by
③ Several usage of chest AI : 1stReader
Diagnosis Normal
classified as normal by computer, radiologists may not interpret the image (or check briefly)
✔Value hypothesis AI
radiologist Abnormal
✔Value hypothesis
- Doctors could focus on potential abnormal cases It’s significant reduce of doctors’ labor
※Regulatory must be clarified for this usage It’s same as other usages, but must be verified carefully
■To improve the environment of chest x-ray diagnosis is Take home message #1
y g one of Konica Minolta’s high priority projects on AI
Focus on Primary Care
■Line-up of Konica Minolta Products for primary care ■ Create movie-like images with use of pulse x-ray.
■ X-ray condition (ex.)
About Dynamic X-ray
Pulse X-ray
・95 kV ・15 flame/sec ・10～20 sec ・pixel：0.4mm×0.4mm
■ Dose
・7 uGy/flame
Upright Position
y Device
Continuous pulse x-ray ・7.3 uGy/flame
In case 15 sec., 7.3 uGy×15 flame×15 sec =1.66mGy
(6)Dynamic imaging x bone suppression
Original Bone Suppression
From Anatomical to Kinetic
COPD
From Kinetic to Functional Blood Depth Changes
+10mm
Normal Abnormal
(7)Maxima Phase Shift Functional Images
+0.50.5
Normal Abnormal
‐‐0.5
■Primary care is one of main targets for Konica Minolta
– Konica Minolta has products
Take home message #2 Konica Minolta has products
• Flat Panel Detector (including dynamic imaging) • Ultrasound
• PACS/WS
• AI(currently in Japan only)
– Computer-aided Detection for Mammogram – Bone Suppression Images for Chest X-ray – Others
hinhanhykhoa.com