Beating Cancer With Two Words: Artificial Intelligence

8 min readApr 5, 2020

“86 percent of mistakes in the healthcare industry are purely administrative and preventable.”

The healthcare industry is always happy to collaborate with technology to develop faster and accurate outcomes. Some of the trends that will disrupt the medical sector are medical imaging, communication between doctors, patients and healthcare centres as well as error reduction. In this article, I will walk you through what’s being done today to improve healthcare, a brief introduction on implementations of AI as well as the challenges and future possibilities of technology and science.

Highlights

AI + healthcare in the present
Companies involved in AI + cancer treatment
Breast cancer algorithm
Support Vectore Machines (SVM’s)
Obstacles with AI + healthcare
Future of AI in healthcare

Keywords

Before I dive deeper into the current and future processes of the ways AI is changing healthcare and how we can leverage it to help classify whether a tumour as cancerous or not, take a moment to familiarize yourself with some key terms I will be using throughout this article.

Artificial Intelligence: machines that act and think like humans

Machine Learning: recognizing patterns and training data to be able to make predictions

Deep Learning: a subset of machine learning using neural networks to train large chunks of data

Biopsy: medical test where cells are removed from a specific area to check for abnormalities

Malignant: a tumour that is made up of cancer cells that multiply at a fast rate.

Benign: tumours that are not cancerous and do not spread to other areas in the body or its tissues.

Status Quo

The great thing about artificial intelligence is its ability to become familiar with patterns from a set of complex data that humans don’t even have the ability to understand. According to Business Insider Intelligence, spending on AI in healthcare is estimated to grow 48% between 2017 and 2023. Currently, AI can detect diseases more accurately than doctors. However, it is still in its early stages. With AI integrated in the health care industry, it can improve:

Speed & accuracy of diagnosis
Clinical decision-making
Faster & better results

Cancer

Cancer treatment is easier during its early stages. It’s great that we have all this technology to gather high-quality images to look for abnormalities, but then comes the time to be hopeful, where the doctor tells you whether the tumour is malignant or benign. Most of the time, a biopsy is run, (most accurate way to search for cancer) but even then, it’s not the most reliable way as it can suggest treatment is needed for those who don’t have cancer and vice versa. Despite, AI even allows for mammograms to run 30 times faster as well as being accurate 99% of the time.

But, AI could potentially change this!

Currently, AI has proven to be successful in cancer diagnostics like radiology and pathology. For instance, radiologists (medical doctors that provide diagnosis and treatment) are really busy people and making a mistake will create a riffle effect of problems. Some of the ways in which AI makes their lives easier is…

Providing a new perspective; thinking about treatment and diagnosis that are different suggested by AI.
Lending a third hand; can takeover monotonous tasks for radiologists

What Companies Are Doing…

Believe it or not, AI is impacting the health sector in numerous ways in which it would take doctors much longer to detect and diagnose. Different companies and startups are doing different things about implementing AI in the healthcare industry.

Google DeepMind Health

One example is Google’s DeepMind Health venture in AI. By working with doctors and researchers, It can detect breast cancer in mammograms more accurately a doctor’s naked eye! With the combination of machine learning and systems neuroscience, it can create algorithms that function like the human brain.

According to Northwestern University, enclosed in the yellow box shows where AI systems have detected cancer within breast tissues where 6 radiologists failed to detect.

“The goal is finding more cancers that are actually going to kill people.” But the problem is “there’s no gold standard for what constitutes cancer.”
— Adewole Adamson, a dermatologist and assistant professor at Dell Medical School.

IBM

IBM Watson created a genomic product that was able to treat more than 1000 patients by identifying a specific treatment. By analyzing massive data, it was able to recommend treatments for patients who had cancer. Essentially, it’s like storing large amounts of data for symptoms, treatments and patients’ medical records and training the data to familiarize itself on how to treat future patients with similar medical issues.

DeepMact

Over 90% of cancer patients die due to distal metastasis (spreading of cancer cells through blood or the lymph system to other areas in the body). Deep learning based Metastasis Analysis in Cleared Tissue (DeepMact) is an algorithm founded by researchers at the Technical University of Munich, the Helmholtz Zentrum München and the LMU Munich “automatically detects metastases”. It consists of two steps:

Mouse bodies become transparent by using a “tissue-clearing” method known as vDISCO and then imaged
Deep learning algorithms are used to detect cancer metastasis

Metastasis can be seen throughout the body through deepMact

Malignant or Benign?

I recently worked on a project using Pycharm to write an algorithm that can determine whether a tumour is malignant or benign. *This was the reason that inspired me to write this article. I will briefly walk you through the lines of code (this is assuming you already have some basic knowledge on python).

First, I import all the necessary data, which includes data from sklearn along with the dataset for breast cancer and support vector machine [cancer = datasets.load_breast_cancer()]*You can scroll down below to read a brief on how SVM’s work. The purpose of the SVM is to classify whether the data is cancerous or non cancerous.

After loading all the data required, I printed out the data for features and labels [mainly since I’m a visual person :)].

Then I split the data into training and testing data and printed out the results because I am very much a visual learner :).

Classifying data into training and testing

Full Code:

import sklearn
from sklearn import datasets
from sklearn import svm

cancer = datasets.load_breast_cancer()

print(cancer.feature_names)
print(cancer.target_names)

x = cancer.data
y = cancer.target

x_train, x_test, y_train, y_test = sklearn.model_selection.train_test_split(x, y, test_size=0.2)

print(x_train, y_train)

Support Vector Machines

A SVM’s purpose is to classify data. A straight line is drawn to divide the data in the way where it has the greatest margin (distance from the line to the nearest point in either sides). This is known as a hyperplane. *Note that SVM’s can work with high dimensional data. In 2D space, the hyper-plane is drawn as a line, whereas in 3D space, it is drawn as a plane.

But what if we have a complex set of data like this…

Kernels

How would you be able to classify it? This is where we can introduce kernels. A kernel is simply a function that takes in the data from the 2D plane and converts it to 3D plane having 3D coordinates. Some examples of the different types of kernels used are…

Linear
Circular
Polynomial
Sigmoid

Using a Kernel to convert 2D data to have 3D coordinates

Obstacles

There are many open holes with sharing data and people’s ethical beliefs in relying on some machine to tell them if they’re sick or not. Some of the major obstacles include…

Black Box

While AI and DL can impact healthcare, there are problems we face, as well. For example, the “black box” problem, where humans don’t know which features AI uses to create an output. In other words, imagine that an algorithm has determined a treatment for a patient, but we don’t know why or how it has come to that conclusion. Why is this important? By knowing the why and how, researchers can understand and discover other features in treating a patient as well as eliminate (if necessary) any biases.

Trust

Other challenges include the fact that not a lot of doctors or patients around the world trust the works of AI to tell them something is or isn’t wrong. However, the purpose of AI isn’t to solely produce an outcome and we must follow along with it, it is there to provide a new perspective to help doctors out with what they can’t notice with their naked eyes.

Liability

With trust comes liability. For AI to be accurate in detecting abnormalities, it would require lots and lots of data to train on. This is where the privacy of patients becomes a major issue. Data sharing transactions occur between healthcare systems and AI companies. This poses ethical questions such as who owns the right to get a hold of one’s data? How secure will patients’ data be in the hands of a 3rd party? What might the consequences be if one’s data is exploited? This could heavily slow down the impact and implementation of AI in the healthcare sector.

What’s Next?

AI is already changing the game for doctors, researchers, patients…. The power and capabilities of how AI can create an impact is so high that it is an estimated to be a $6 billion industry by 2021. It is expected to become more accurate in the diagnosis and treatment of patients. Experts hope to minimize errors to improve operational efficiency and accuracy. First and foremost, there are many challenges to deal with along the way such as biases within AI, trust and liability… but if we can get passed this, the AI will help save millions and billions of lives.

Key Takeaways:

AI has the ability to predict patterns and function the way humans think and understand
AI + healthcare can generate a new perspective of treatments for patients faster and more accurately
AI is used in healthcare, for example in radiology
There are many companies that work with clinicians and researchers to collect more data and have their algorithms produce accurate results
SVM’s are used to classify higher dimensional sets of data
Kernels converts data from a 2D plane to a 3D plane
There are problems with AI playing a role in healthcare like machines generating biases, lack of trust for doctors and researches as well as who is responsible if a security breach occurs.
AI will heavily impact the future, but as of now, there are many challenges to overcome.

Resourceful Links

If you enjoyed this article, you can support, follow and subscribe to my social media! :)

LinkedIn: www.linkedin.com/in/tanyayousofi
Youtube: https://www.youtube.com/channel/UCxDucQ357qJiXsAN_1ybOrQ