2. WHY TO USE CNN?
• Supply of doctors is limited in developing
countries like India, especially in smaller
towns and villages making provision of
healthcare is difficult to a larger group of
people.
• Cost of skin cancer detection is approx. 4000
INR / 400 TL / $55
• If Skin cancer is not detected in early stage, it
cannot be cured.
• Now in the age of digital assistants like
Google and Alexa, can we create cancer
detection tools using CNN that can diagnose
diseases like cancer at an early stage.
3. 1. Research indicates that most experienced physicians can
diagnose cancer with 79 percent accuracy while 91 percent
correct diagnosis is achieved using machine learning techniques. In this
case study, our task is to detect 3 most common types of cancers at an
early stage using CNN.
3. There are mainly 3 types of cancers:
Basal-skin cell cancer
Squamous-skin cell cancer
Melanoma cancer
INTRODUCTION
4. WHAT IS CNN ????
In Deep Learning, a Convolutional Neural Network
(CNN) is inspired by biological processes in which the
connectivity pattern between neurons resembles the
organization of the animal visual cortex.
It is an AI function that mimics the workings of the
human brain in processing data for use in detecting
objects, recognizing speech, translating languages, and
making decisions. Deep learning AI is able
to learn without human supervision.
6. 1. An image of 28x28 pixels
Representing a handwritten digit
9.
28x28 = 784 neurons needed in
1st layer.
2. Each neuron represents the
value of pixel ranging from 0 to 1.
Value 0 = neuron not activated
Value > threshold = active
8. Hidden layers:
Doing all the
complex
computation work
inside the neural
network
What are these layers in between of the network?
9. How does an Artificial Neural Network recognize patterns?
Pretty same way
Every neuron in a
hidden layer
can recognize the
hidden patterns in an
image by
subcomponents
Inside each neuron.
Activation of that
particular neuron
will be close to 1
10.
11. Each Hidden layer / Convolution layer works by using
filters that detect the patterns in the image.
12. Activation function in Neural
Network
activation function is
a function that is added into an
artificial neural network in order to
help the network learn complex
patterns in the data. When
comparing with a neuron-based
model that is in our brains,
the activation function is at the
end deciding what is to be fired to
the next neuron.
13.
14. Convolution layer: Here, the filter traverses throughout the
image to find matches. The math performed here is called a
convolution
Pooling Layer: Pooling is a way to take
large images and shrink them down while
preserving the most important information
in them.
1080*1080 = 1166400
18. Dataset
1. Images are taken from
the HAM10000 dataset.
This dataset is a
collection of multi-
source dermatoscopic
images of common
pigmented skin lesions.
2. The dataset is split into
two parts, 70% for
training and 30% for
testing.
3. Transfer learning is
applied on to models
19. STRATEGY
Skin Scan with Mobile (having application
Trained using CNN)
Cancer identified
Symptom assessment using Deep learning
Convolutional Neural Network
Report
20. Approach we used
ImageNet:
Neural network
developed by
Stanford
University trained
over 14,000,000
images.
Train last layer to
recognize custom
classes.
22. UNIQUENESS
• We will be using Convolutional Neural Network for early stage cancer
detection.
• We will also be using Cloud Computing for faster processing and reduce
system overload.
• This automatic system will reduce human effort and costs with improved
accuracy.
TECHNOLOGIES USED
•Tensorflow
•Convolutional Neural Network
•Deep Learning
•Cloud
•Android
24. Conclusion
The transfer learning techniques helps us to reduce the training time by using the pre-trained
weights from previous model. It also does not requires much data to train as well boosting performance of the
neural network.
Modified SSD mobilenet v1 coco is a better choice for implementing it in mobile application if speed is
concerned but very poor if accuracy is to be concerned. The used models follows CNN approach to produce
better detection accuracy.
In our case, accuracy is the only factor because for cancer detection therefore, false positives and false
negatives are minimized. In future, it can be can be implemented as a composite model that can detect other
types of cancers such as mouth cancer and other forms of cancers.
Editor's Notes
We piece together many componentss
1 Neuron corresponding to every pixel
There is some specific neuron that will have activation close to 1
1st hidden layer lights up all the neurons that are associated with those little 8-10 little edges
Which inturns lights up 2nd hidden layer