Classifying Skin Lesions with Convolutional Neural Networks

With this in mind, I set out to make an end-to-end solution to classify skin lesions using deep learning.A way that we can make accurate and reliable medical image analysis tech is through the use of Convolutional Neural Networks — a type of deep neural network that is used to analyze images..Before we get into CNN’s, let’s backtrack a little bit and take a look at how neural networks work..Neural networks are essentially computer programs that are modelled in a similar fashion as to how a human brain works..Like our brain, neural networks consist of a whole bunch of neurons that can’t do much individually, but when connected together in a network, can do some pretty incredible tasks.What does a neuron do?A neuron takes in some inputs, operates on it, and gives an output. Source.The job of a neuron is quite simple, it takes in a couple inputs; let’s call them X1, X2, and X3, as per the diagram, and then spits out an output..Each of the inputs has a certain weight coefficient associated with them..These weights affect the output of the neuron and they are constantly changed to get an improved output(that’s more accurate).Think of a neural network as a colony of termites..A termite by itself is insignificant and can’t really do anything, but when you have millions of them all together, they can do incredible tasks by learning and eventually evolving.A neural network with 2 hidden layers. Source.Neural networks are often used as a method of supervised learning, which basically means that they learn how to make predictions by recognizing patterns in data..The more data we give it, the better it will be at making predictions..But how does a neural network find patterns in some of the really complex data that we work with?It learns in a similar way to how humans learn..At first, it goes through the data and makes super random predictions that aren’t very accurate at all, then it uses a technique called backpropagation, where it learns from the mistakes it made then tries to fix them..The more it does this, the better it gets at predicting..They often can even recognize patterns and correlations that a human could never see!So how do we apply this to pictures of skin lesions?. More details

Leave a Reply