How Neural Networks Will Save Millions Of People

As I said lung cancer is usually diagnosed when the cancer is in its later stages making it hard to treat. Walter White from the Breaking Bad series is a good example, my favorite show😎. He was diagnosed with lung cancer when it was in its later stages, making it inoperable.

So why is lung cancer detected so dang late? Well, the main reason is that the process of detecting lung cancer early is pretty hard. To get a diagnosis for early lung cancer you have to get a CT(computed tomography) scan which has 200–300 images. After the CT scan, a doctor will look at the scan and check to see if any of the lung nodules from the CT scan are cancerous, then the patient will get the results after about a couple of stressful days.

Lung nodules are abnormal growth in your lung; usually, lung nodules aren’t cancerous but when they are they can be easily missed especially if the cancerous lung nodule is very small🤏. When you don’t treat a cancerous lung nodule it will continue to grow and spread then it gets found and becomes really difficult to treat effectively.

In the image below you can see the little white dots circled are cancerous. You can see how similar they are to all the other dots around them. Now imagine looking through 200 images of tiny dots and trying to figure out if any of the dots are cancerous. That’s very hard and even fully trained pulmonologists will make mistakes. Once a cancerous lung nodule goes undiagnosed the cancer will continue to quietly grow and spread to other parts of the body.

“Wow, lung cancer is very hard to find early… I wish there was a better way to diagnose it!” This is where the neural network comes in.😎

Neural networks are a very complex form of AI🤖 and are used to learn almost anything. For now, I will explain the simplest neural network called “feedforward neural network”, aka FFNN or just a neural network. After this, I will explain how neural networks can detect cancerous lung nodules.

Neural networks are made up of layers, the input layer, hidden layers, and the output layer. The first layer in a neural network is the input layer which has a bunch of neurons. Neurons sound fancy but all they do is hold a value, in the video above the circles are neurons. In this example, the image of the number seven is being fed into the neural network. This means each neuron in the input layer will have a pixel’s greyscale value. Also in the first layer, there are 784 neurons or circles, this is simply because the image has 784 pixels so each neuron has one pixel’s greyscale value. Greyscale value is just how black or how white the pixel is. 1 being black 0 being white and 0.5 being gray

The next layer in a neural network is called the hidden layer. The hidden layers are supposed to find patterns in the data it is trained on to maximize accuracy. Hidden layers find patterns in data using complex math equations intended to get the right answer, like activation functions, weights, and biases. But that is too complicated and you don’t really need to know about how it works. Put more simply, the hidden layers are just finding common patterns that the input has, like how the number zero is shaped like a circle or how a seven has a line on the top and a diagonal line going down.

Finally, the output layer is the answer that the neural network is most confident with. So that means if the neural network is 70% sure that the number is a four and 30% sure that the number is a five it will choose the output it is most confident which in this case is the number four.

So, in summary, a neural network has three different types of layers the input layer which has all the input data, hidden layers which narrow down what the output will be, and the output layer which will choose what the output is. Keep in mind this is a very simplistic explanation of how a neural network works and in reality, neural networks have a lot of math involved in how they learn and find patterns so I will have some resources at the bottom of the page for more in-depth explanations.

Now that we know how a neural network operates how does it find cancer? Well, the neural network will be trained on tons of images of ct scans of lungs that have cancer and those that don’t, so the neural network will learn what a cancerous lung nodule looks like and what a noncancerous nodule looks like. The neural network that will be most useful would be a convolutional neural network(CNN) which is more complicated than a feedforward neural network. The neural network aims to assist a pulmonologist in their diagnosis.

Also, I would like to say that to become a pulmonologist you have to take four years in college, complete four years in medical school, and then do a three-year residency program. While the neural network would only take hours or maybe days to train with accuracy.

Alright, so we both know that this type of technology is possible, so what companies are using this technology?

Companies that are using neural networks to detect lung cancer are Google and Nvidia.

Google made a neural network to detect lung cancer and is available via the google cloud healthcare API. Google’s AI outperformed six radiologists that had an average of eight years of experience. Many companies like Johnson & Johnson, Facebook, and Intel have used the google cloud API.

Another company that has a neural network to detect lung cancer is Nvidia. Nvidia is using their GPUs to train neural networks that are being used to detect lung cancer an example is 12 sigma technologies which are detecting lung cancer a video about them will be at the bottom of the page.

We know how the neural network works and how it detects lung cancer but why should we trust it more than a trained doctor do all the work? One example is Google. Google claims that its AI outperformed six radiologists with an average of eight years of experience. The AI detected cancer 5% more accurately. And 11% more likely to reduce false positives (aka misdiagnosis). Neural networks are also way more efficient than radiologists what may take up to a couple of days to get back a diagnosis will take less than an hour for a neural network to get a diagnosis.

All in all the main benefit of using neural networks is not to replace the pulmonologist but to work with them to get a more accurate diagnosis quicker(especially in lung cancer early stages).

Related posts:

American Progress is a distinctly American painting by John Gast, envisioning American expansion across the North American continent. The image portrays a woman clad in white, flying over groups of…

Throw on this cotton t-shirt when you want a casual but unique outfit these short sleeve t-shirts will make people stop in their tracks to read the graphics that youre sporting, capturing the…

Watched so many posts of anger today. Decided to check out. Spent no time on social media after about 8am and you know what? I had a delightfully happy day. Also today I got to escort two puppies to…