1 of 11

Where can image classification be used?

  • Medical imaging is a critical part of the healthcare system

  • As AI becomes more advanced, it will start playing more of a role in diagnostics and therapeutic care for patients

2 of 11

Important Problem: Cancer

3 of 11

How does AI detect Cancer Cells

This is a much harder problem than identifying animals or plants

  • Need to image cells, which don’t feature as much natural variation as do larger objects

  • Doctors spend years of training to identify cancer cells

  • Can/should we train AI algorithms to do it?

4 of 11

What is Mitosis

Mitosis is cell division

  • one cell divides to produce two identical “daughter” cells
  • this is done in an organized way, to make sure that the two cells created at the end have exactly the right number of chromosomes

Why is it important?

  • to grow, for example a baby growing into an adult
  • to regenerate, for example a wound healing
  • to reproduce, in single-celled organisms like yeast this is how they reproduce

The phases of mitosis

  • Prophase
  • Metaphase
  • Anaphase
  • Telophase

5 of 11

What is the normal cell cycle

Normally, cells go through a typical cell cycle which involves several phases

1 - Interphase

    • first the cell has to grow
    • then the cell has to copy its DNA (genetic material)
    • the cell then grows some more to prepare for division

2 - Mitosis phases

    • the cell divides to make two identical daughter cells
    • there are 4 stages of Mitosis.

3 - Prophase

    • spindle fibers form and chromosomes pair up

4 - Metaphase

    • Spindle fibers pull the chromosomes to the center of the nuclei

5 - Anaphase

    • spindle fibers begin to pull the chromosomes apart
    • 6 - Telophase
    • two new nuclei begin to form

6 of 11

Example Cell Cycle

chromosomes

nuclei

spindle fibers

Interphase

Prophase

metaphase

anaphase

telophase

cytokinesis

1

2

3

4

6

5

7 of 11

What is Cancer

  • Uncontrolled Mitosis
    • many cells continue to grow and divide
  • Cancer cells, grow without input from the body.
    • large dark nucleus
    • can grow their own blood supply
    • abnormal number of chromosomes, unorganized
  • Cancer cells, ignore signals that would cause them to stop dividing.
    • variation in size and shape
  • Cancer cells don’t age or die like a normal cell.

For the purposes of this vignette

we will define “non-mitosis” as cancer.

8 of 11

How can AI help us?

  • Dataset of breast cancer histological images

  • Task: determine if cell is undergoing mitosis or non-mitosis

  • How accurate are humans? How accurate are medical doctors?

9 of 11

Teachable Machine: Mitosis detection

  • Let’s use a neural network to do image classification

  • First collect a dataset of training images: mitosis and non-mitosis

https://drive.google.com/file/d/1ffEVJo7tKtQopNA1O9LMSRpqgk9x8RaW/view?usp=sharing

  • Train a model on Teachable machine to classify between the two classes

  • Test the model on test data (not seen before by the neural network)

10 of 11

Questions for Conversation

Will this AI work if the cells are presented under a different reagent? or at a different magnification?

If this AI is accurate 80 percent of the time in this application, what is it’s value to science?

Are there ethical use questions for this type of data, and does that affect the outcome of our database, and AI?

11 of 11

TikTok Algorithm of MetaData vs a Neural Network for Images

  • Tiktok gets data from tags, likes, time watched, what other people like - and uses this data to recommend a video.

  • TikTok algorithm does not see the video or recognize the content of the video

  • A Neural Network (like the teachable machine) �can learn to see the differences in the images�presented to it.

  • This AI won’t be accurate 100 percent of the time, �but can be used to assist trained professionals