Skin cancer Detection

Problem statement:

Skin cancer is the most diagnosed cancer all over the world, It greatly affects one's daily life. Skin cancer also causes body dysmorphia.The objective of the skin cancer detection project is to develop a use case to analyze and assess whether the cancer is benign or malignant so that individuals can take action immediately.

Usage domains and advantages:

The medical industry can use this model for identifying potential patients' cancer cells stage as benign or malignant.

Individuals can check their lesion condition using this model.

By understanding which stages their lesion is one can take precautions to avoid the risk.

Skin cancer detection using machine learning is known as intelligent computational predictive systems. These are proven to be very effective in many medical industries.

Model solution:

The problem is about detecting a person's skin lesion stage whether it is benign or malignant.

To identify these lesion stages using images of these stages a Sequential model is considered to be more suitable.

Dataset/Data Source:

Dataset is from Kaggle healthcare and medical datasets

https://www.kaggle.com/fanconic/skin-cancer-malignant-vs-benign

Dataset is divided into test and train sets which are further divided into benign and malignant

Train set contains 2806 images of both benign and malignant

The test set contains 304 images of both benign and malignant

Images shape 224, 224

Width 224, Height 224

Preprocessing:

All images are in the shape of 224x224 width and height

Then are explicitly converted to RGB for getting channel as 3

Separate labels are created for both benign and malignant

There are divided to train and test NumPy arrays

Shapes: 

X_train:  (2806, 224, 224, 3)  

Y_train:  (2806,)  

X_test :  (304, 224, 224, 3)  

Y_test :  (304,)  

X_train - Benign and Malignant images

Y_train - labels 

X_test - Benign and Malignant images

Y_test - labels 

Model Used:

Keras - Sequential model

Model additional layers 

• Convolution with input activation as ‘relu’


• Max pooling with (2,2) kernel size


• Dense output layer with ‘softmax’ activation

The optimizer used is ‘adam’,

The loss function is ‘categorical_crossentropy’

Metrics used is ‘accuracy’

Results:

Below are the results yielded by the model.