Making Object Detection Work in Medical Imaging Idan Bassuk - - PowerPoint PPT Presentation

▶
making object detection work in medical imaging
SMART_READER_LITE
LIVE PREVIEW

Making Object Detection Work in Medical Imaging Idan Bassuk - - PowerPoint PPT Presentation

May 17, 2023 168 likes •573 views

Making Object Detection Work in Medical Imaging Idan Bassuk Segmentation Detection Classification 4 5 Mature technology 6 Mature, Accurate and Fast 7 Changing the world, one killer app at a time 8 Changing the world, one killer app at

slide-1
SLIDE 1
slide-2
SLIDE 2

Making Object Detection Work in Medical Imaging

Idan Bassuk

slide-3
SLIDE 3

Classification Detection Segmentation

slide-4
SLIDE 4

4

slide-5
SLIDE 5

5

slide-6
SLIDE 6

6

Mature technology

slide-7
SLIDE 7

7

Mature, Accurate and Fast

slide-8
SLIDE 8

8

Changing the world, one killer app at a time

slide-9
SLIDE 9

9

Changing the world, one killer app at a time Visual Intelligence (Military and Business )

slide-10
SLIDE 10

10

Changing the world, one killer app at a time Robotics

slide-11
SLIDE 11

11

Changing the world, one killer app at a time Medical Imaging is Exploding

slide-12
SLIDE 12

12

Changing the world, one killer app at a time Medical Imaging

slide-13
SLIDE 13

First - Understanding the Algorithms

slide-14
SLIDE 14

14

What is a Convolutional Feature Extractor?

slide-15
SLIDE 15

15

Convolutional Feature Extractor

slide-16
SLIDE 16

16

Person 97%

Convolutional Feature Extractor – Plug and Play

slide-17
SLIDE 17

17

C1 97% C2 2.9% C3 0.1%

What is a classification head?

slide-18
SLIDE 18

18

x 12 px y -10 px w 42 px h 50 px

How to plug in regression?

slide-19
SLIDE 19

19 Human Level

(Approximate)

Before Deep Learning

Classification accuracy

slide-20
SLIDE 20

20

How to REALLY Plug in Regression? Sliding Window

Bounding Box Coordinates BG 90%

slide-21
SLIDE 21

21

How to REALLY Plug in Regression? Sliding Window

Bounding Box Coordinates BG 93%

slide-22
SLIDE 22

22

How to REALLY Plug in Regression? Sliding Window

Bounding Box Coordinates

Person 97%

slide-23
SLIDE 23

23

How to REALLY Plug in Regression? Sliding Window

Bounding Box Coordinates

Unicycle Wheel 95%

slide-24
SLIDE 24

24

The Single-Stage Detection Algorithm

Bounding Box Coordinates x=5, y=-2, w=62,h=66

* SSD, YOLO 9000

Unicycle Wheel 95%

slide-25
SLIDE 25

Guidelines for Adapting Object Detection to Medical Imaging

slide-26
SLIDE 26

#1: Question the basic assumptions

slide-27
SLIDE 27

Architecture Meta-Architecture Post Processing Detection without Pre-Training Focal Loss Soft NMS Deformable Convolutions Multi-Task Learning Learned NMS Deformable ROI-Cropping Feature Pyramid Networks

#2: Being on top of the research

slide-28
SLIDE 28

#2: Being on top of the research Solving Scarcity of Positive Data

slide-29
SLIDE 29

#2: Being on top of the research Attention is Different

slide-30
SLIDE 30

30

#3: Solving Detection in a Large 3D Volume

Images of cats 1024x1024x3 Cat scans (CT) 512x512x300

input size x30 times larger

  • bject size x10 times smaller
slide-31
SLIDE 31

31

#4: Polyglot Data

Patient data Scan Diagnosis

  • 1. Demographics
  • 2. Referral Letter
  • 3. Past Scans & Reports
  • 4. Scanner Meta-data
slide-32
SLIDE 32

Key Takeaways

slide-33
SLIDE 33

Key takeaways

1. Medical Imaging is one of the quickest growing bottlenecks in the medical world

slide-34
SLIDE 34

Key takeaways

  • 1. Medical Imaging is one of the quickest growing bottlenecks in the

medical world

  • 2. Object Detection is a mature technology with many

promising applications in Medical Imaging

slide-35
SLIDE 35

Key takeaways

  • 1. Medical Imaging is one of the quickest growing bottlenecks in the

medical world

  • 2. Object Detection is a mature technology with many promising

applications in Medical Imaging

  • 3. Adaptation is the name of the game
  • 4. It’s hard - but achievable
slide-36
SLIDE 36

Thank you!

Idan Bassuk idan@aidoc.com

slide-37
SLIDE 37

#1: Being on top of the research

The Basics - a. Faster R-CNN b. Single Shot Detector (SSD) c. R-FCN d. YOLO , YOLO 9,000 e. Speed/accuracy trade-offs for modern convolutional object detectors f. Tensorflow Object Detection API

slide-38
SLIDE 38

Advanced Material: a. Fast R-CNN b. Deep Learning and Object Detection Tutorial by Ross Girshick and Kaiming He c. Focal Loss for Dense Object Detection d. Deep Residual Learning for Image Recognition e. Feature Pyramid Networks for Object Detection f. Beyond Skip Connections: Top-Down Modulation for Object Detection g. DSOD: Learning Deeply Supervised Object Detectors from Scratch h. Instance-aware Semastntic Segmentation via Multi-task Network Cascades i. Fully Convolutional Inance-aware Semantic Segmentation j. Deformable Convolutional Networks k. Mask R-CNN

#1: Being on top of the research

slide-39
SLIDE 39

1) Idan Bassuk - idan@aidoc.com 2) TED.COM - Joseph Redmon: How a Computer Learns to Recognize objects Instantly 3) Satellite Image 4) Facebook.com 5) Cloud Factory 6) Real-Time Grasp Detection Using Convolutional Neural Networks 7) ImageNet Classification with Deep Convolutional Neural Networks 8) Visualizing and Understanding Convolutional Networks 9) http://www.asimovinstitute.org - The Neural Network Zoo 10) You Only Look Once: Unified, Real-Time Object Detection 11) Pokemon Go - vg247.com 12) Neurala

Credits for graphics