How Applying AI to Cancer Can Save Millions

I can almost guarantee that every single one of you reading this article knows someone who has battled with cancer. It sucks. Recently Chadwick Boseman, the legendary actor who played Black Panther, shocked the world when he tragically passed away from colon cancer.

I mean, seriously. The man played a superhero in a movie — HE LITERALLY LOST A FIGHT AND GOT THROWN OFF A HUGE MOUNTAIN AND STILL SURVIVED IN THE FREEZING COLD, and yet cancer still was able to take him out in the real world. WAKANDA FOREVER ❤️.

According to the National Cancer Institute, roughly 40% of men and women will be diagnosed with cancer at some point in their lifetimes. In 2020, it is estimated that 1,806,590 people will be diagnosed with cancer for the first time in their life, and more than a third of them will pass away. That’s crazy. With modern technology, there must be ways to change that.

However, before we jump straight into trying to find the cure for cancer, we have to at least have a basic understanding of the cell cycle. Then we can dive deeper into what cancer is and how it arises in the body. Stick around to learn more.

Cell Division

All humans start off as a single cell: a fertilized egg cell. First of all, that’s crazy to even think about — think about how far you’ve come! But on a serious note, how did that one cell become who you are today.

As you may know, the cell divides. Cells divide for a multitude of reasons, and here are the main ones.

  1. Growth and Development
  2. To Repair/Replace Damaged Cells
  3. To Stay Small for Efficient Nutrient Exchange
  4. To Reproduce

How Do Cells Divide?

Before they divide, cells actually spend most of their time in the different stages of interphase, the resting time between successive mitotic divisions. There are three phases of interphase: G1, S, and G2.

A diagram that illustrates the different stages of the cell cycle

G stands for growth and S stands for synthesis. In the G1 phase, the cell does most of its living, growing, and functional responsibilities. In the S phase, the DNA is replicated so that when the cell actually divides, each daughter cell has an exactly identical number of chromosomes as the parent cell. Finally, there’s the G2 phase, which is all about creating any other organelles or molecules to prepare for cell division in mitosis.

Diagram of the cell and its chromosomes while in interphase. As we move through the stages of mitosis, we can see how this changes.

Next up in the cell cycle is the splitting of the nucleus, also known as mitosis. Mitosis is split into four stages: prophase, metaphase, anaphase, and telophase.

In the left-most illustration, the cell is in the G1 phase of interphase. That is a homologous pair of unreplicated chromosomes. They replicate in the S phase, and then the next image in the diagram is in prophase, then metaphase, then anaphase, then telophase, and finally cytokinesis.
  1. Prophase = the DNA and proteins condense and worm-like figures slowly become visible
  2. Metaphase = the replicated DNA lines up in the middle
  3. Anaphase = spindle fibers pull the sister chromatids to opposite sides of the cell
  4. Telophase = sister chromatids reach the opposite poles of the cell, DNA starts to uncoil to form chromatin, and the middle of the cell starts to split (either a cleavage furrow or a cell plate)
Left: Animal cells dividing by use of a cleavage furrow and Right: plant cells divide using a cell plate. Plant cells cannot divide like animal cells because plant cells have cell walls.
Top Left: Prophase, Bottom Left: Metaphase, Top Right: Anaphase, Bottom Right: Telophase

Phew! Now we know how healthy cells are supposed to divide and behave, so how does this have any correlation to cancer?

What Even is Cancer?

Cancer cells kinda work like old faucets. Sometimes when you try to turn off the faucet, it doesn’t work; consequently, water keeps on flowing, and eventually, there’s so much water and your house floods. Similarly, cancer cells ignore the body’s signals to undergo apoptosis or programmed cell death. Apoptosis is the body’s way of removing old cells with new, healthy cells. Instead, cancer cells keep on dividing and eventually form a solid clump of cancer cells, also known as a tumor.

Most cancers form solid tumors, but blood cancers, such as Leukemia, do not. In this article, we will mainly be focused on cancers that form solid tumors because they make up the majority of the types of cancer.

There are 2 types of solid tumors:

  1. Benign tumors are not harmful or infectious (unless it is in the brain, then it can be life-threatening), and when removed, they do not grow back.
  2. Malignant tumors, on the other hand, are extremely harmful and infectious. Once the tumor gets to the size of a grain of sugar, it uses angiogenesis, which sends a chemical signal towards nearby blood vessels, causing them to sprout and grow towards the tumor. Once connected, the tumor now has a supply of oxygen and nutrients which allows the cancer cells to divide more and cause more harm to the body.

In addition to providing the tumor with necessities, the blood vessels also act as transportation for the cancer cells. Some cancer cells break off from the tumor and spread to other parts of the body via the bloodstream, potentially creating more tumors — this process is known as metastasis.

Once a tumor grows too big, it starts to invade the space surrounding healthy tissues and organs. These tissues and organs need that space for nutrients. Without them, they fail to function, causing the patient to die.

What Causes Cancer?

Normally, healthy cells grow and divide in order to fulfill the body’s needs. When a cell is old and damaged, it undergoes apoptosis and a healthy cell takes its place. So how do healthy cells turn into cancer cells?

Cancer is a genetic disease — this means that healthy cells turn into cancer cells through mutations in the cell’s DNA. Mutations can occur in two ways: as a cell is dividing, a random mistake in the process can cause a mutation or it can also be caused by environmental carcinogens, such as tobacco smoke.

To help spread awareness on carcinogens, the International Agency for Research on Cancer (IARC) and the US National Toxicology Program (NTP) have published a list of substances that can cause cancer.

Genetic mutations that lead to cancer can affect three types of genes, known as the drivers of cancer.

  1. The positive regulators — in healthy cells, proto-oncogenes instruct proteins to help stimulate cell division and delay apoptosis. However, when a mutation occurs in a proto-oncogene, it becomes a cancer-causing oncogene. Oncogenes have mutations in the amino acid sequence in their proteins, which essentially leaves the “divide” stimulator always on and can lead to a tumor.
  2. The negative regulators— tumor suppressor genes pretty self-explanatory. They suppress the formation of tumors. The most notable tumor protein p53. When it detects damage in a cell’s DNA, p53 halts the cell process and buys time for DNA repair. If that doesn’t work, p53 has its own set of DNA repair enzymes to try and speed up the process. Finally, if all else fails, p53 will activate apoptosis and destroy the cell and the damaged DNA. However, if p53 has a mutation, it can’t function normally and stop the passing of cells with damaged DNA.
  3. The DNA repair genes — if these genes have mutations, well, let’s just say… that sucks. Because now they can’t actually repair the genes. Come on, genes! YOU HAD ONE JOB!

The National Cancer Institute also discovered that random mutations are the cause for 60% of all cases of cancer; the other 40% is caused by environmental carcinogens. By being more aware of our surroundings, we can drastically decrease the number of cancer cases, but there’s still that 60%.

Now that we understand the basics of what causes cancer and how it destroys the body, we can work on how to tackle it. Let the party begin!

How Do We Treat Cancer Right Now?

Traditional treatments for cancer include but are not limited to surgery, chemotherapy, and radiation therapy.

Depending on the development of the tumor, surgery can be used to remove or debulk the tumor, and consequently lessen or eliminate the pains caused by the tumor. However, a downside is that surgery only works on solid tumors and tumors that have metastasized.

Radiation therapy uses small doses of radiation to damage the cancer cells’ DNA and ultimately cause the cancer cells to die. However, it may take days or even weeks for the radiation to actually harm enough DNA for the cancer cells to die.

The location that the radiation actually enters the body determines the side effects in each specific case. Usually, radiation therapy causes a minimum of fatigue, hair loss, and skin changes. Read here for a more extensive understanding of radiation therapy side effects.

Chemotherapy, on the other hand, is a treatment of cancer that uses drugs to kill cancer cells. Often, chemotherapy is used in collaboration with other forms of cancer treatment. It can be used to make a tumor smaller prior to surgery or radiation therapy: this is known as neoadjuvant chemotherapy.

Chemotherapy can be used after surgery or radiation therapy. This is adjuvant chemotherapy, and it can kill any leftover cancer cells that remain even after surgery and radiation therapy.

Chemotherapy stunts the growth of rapidly dividing cancer cells, but it also affects healthy cells as well. Side effects of chemotherapy are fatigue, mouth sores, nausea, and hair loss. Fortunately, these get better after chemotherapy is finished.

Other cancer treatments are promising and gathering a lot of hype attention, but we won’t cover them in this article. If you’re interested in learning more, you can click on the name of the treatment and it will take you on your learning adventures: immunotherapy and CRISPR gene editing.

The main problem is that certain treatments may work on certain patients, but it ultimately varies from patient to patient. Accurate examinations have huge impacts on both prognosis and treatment decisions. That’s why efficient and accurate technology is crucial.

In this article, we will go over how we can use AI to change the game of cancer forever. THIS IS HAPPENING RIGHT NOW!!

Artificial Intelligence Involvement in Cancer

LYmph Node Assistant, or LYNA for short, is a convolutional neural network developed by Google AI in October 2018 that can accurately distinguish images with tumors from images without 99% of the time. Convolutional neural networks are specifically good for processing images. For an in-depth explanation of how CNNs work, I highly suggest you read this article.

Essentially, the computer uses a smaller filter of pixels to break down the image into smaller chunks, and in these chunks, we can detect vertical and horizontal edges, which we can use to classify a tumor. Super basic concept, but I highly suggest you read that article.

Example of a Convolutional Neural Network, which takes in as input an image of a vehicle and then classifies and returns it as a car, truck, van, bike, etc.

Without using AI, the gold standard in detecting nodal metastases leading to breast cancer is a pathologist’s examination. However, studies have shown that 25% of examinations would be changed upon a second review, and pathologists can only detect micrometastases at around 38% accuracy. That’s not good!

Right: LYNA sorting through the image, pixel by pixel, and correctly identifying the tumor, highlighted in light blue

AI will not necessarily cure cancer, but it will diagnose tumors faster and far more efficiently, leading to better treatment decisions and ultimately prevent the majority of the harm done by cancer. AI has potential, and in the near future, it can save millions around the world.

This does not mean, however, that AI will directly replace pathologists in hospitals. AI is a supplement to improve human life conditions — it is meant to work in conjunction and synchronously with humans, as neither one is better than the other. That being said, Google has also admitted that LYNA still needs more testing because it has only been tested on two datasets: pathology slides from the Camelyon Challenge and the Naval Medical Center in San Diego.

Table of the datasets

Pro tip 🔑: the more data that a model can be trained on, the more accurate the model will be.

Left: sample view of lymph nodes with poor processing quality. Right: LYNA accurately identifies the tumor (red)

So How Does LYNA Work?

LYNA specializes in image recognition of tumors. It works in a hierarchy of layers, modeled after interactions between neurons in a human’s brain, or a biological neural network. Google AI’s convolutional neural network takes in a 299-pixel image as input (mainly because LYNA is based on Inception-v3, a pre-existing image recognition deep learning model).

Deep learning just means that the neural network is more complex because it has an input layer, an output layer, and more than one layer in between. These middle layers are known as the “hidden layers”.

I definitely recommend watching this video for a super in-depth explanation of how a neural network generally works. It can explain it much better than I can in this short article.

Essentially, each neuron in the network acts like a function, taking in all inputs from the previous layer and spitting out a number from 0 to 1 as an output, which becomes the input for the next layer.

In this case, 0 represents a benign tumor and 1 represents a malignant tumor.

Throughout the network, LYNA makes predictions based on the analysis of the image at the pixel-level — so either benign, malignant, or if it's unsure, somewhere in the middle.

After each image slide, the model’s weights and biases are tweaked. This is important to not only improve after each iteration that an image is fed in but also to be able to measure LYNA’s accuracy as a metric in terms of a loss function. The entire purpose of a neural network is to minimize the loss function and thus increasing the overall accuracy of the model.

In a broader sense, earlier layers help to discern features, such as small edges, and later layers help discern more complex features, such as sides.

Basic step by step process of how to train a neural network. The model predicts the output, and judging by how well it does, it will update its weights and biases accordingly. Then it moves on to the next input and keeps on iterating through the entire input data set.

With enough data and as many necessary incremental changes to the weights and biases, LYNA was able to correctly identify tumors 99% of the time.

While being a major stepping stone into the future, the road to integrating LYNA into a hospital with real patients is a long and strenuous path. For example, in this study, each patient had one slide of their lymph nodes, when pathologists normally must examine multiple slides in order to give a complete examination. That being said, the future appears bright for AI to take center stage in hospitals around the world.

Thanks for Reading!!

Stay tuned for more interesting articles, and if you want to talk with me, my DM’s are always open!

Works Cited

3Blue1Brown. “But What Is a Neural Network? | Deep Learning, Chapter 1.” But What Is a Neural Network? | Deep Learning, Chapter 1. Youtube,
www.youtube.com/watchv=aircAruvnKk&t=1013s&ab_channel=3Blue1Brown. Accessed 16 Nov. 2020.

“Cancer Statistics.” National Cancer Institute, www.cancer.gov/about-cancer/
understanding/statistics. Accessed 16 Nov. 2020.

Couzin-Frankel, Jennifer. Cutting-edge CRISPR gene editing appears safe in three cancer patients. ScienceMag, 6 Feb. 2020, www.sciencemag.org/news/2020/02/ cutting-edge-crispr-gene-editing-appears-safe-three-cancer-patients. Accessed 16 Nov. 2020.

Glenny, Helen. “Google AI better than doctors at detecting breast cancer.”
Science Focus, 16 Nov. 2018, www.sciencefocus.com/news/
google-ai-better-than-doctors-at-detecting-breast-cancer/. Accessed 16 Nov.
2020.

“Juno Therapeutics.” Juno Therapeutics, www.junotherapeutics.com/home/. Accessed 16 Nov. 2020.

“Known and Probable Human Carcinogens.” American Cancer Society, www.cancer.org/cancer/cancer-causes/general-info/known-and-probable-human-carcinogens.html. Accessed 16 Nov. 2020.

Stumpe, Martin. “Applying Deep Learning to Metastatic Breast Cancer Detection.” Google AI Blog, 12 Oct. 2018, ai.googleblog.com/2018/10/
applying-deep-learning-to-metastatic.html. Accessed 16 Nov. 2020.

17 y/o innovator | AI enthusiast