Skin Cancer Classification ML pipeline using CNN & Pre-Trained Models(Transfer Learning)

Overview

• This project focuses on the binary classification of skin cancer images into "Benign" or "Malignant" categories. The implementation includes a Convolutional Neural Network (CNN) and/or Transfer Learning using TensorFlow and Keras.
• The dataset is not provided here, but you can use your own Skin Cancer image dataset.
• I have used 10,000 images for Training .
• 5000 Benign and 5000 Malignant images with Train:Validate:Test ratio 7:2:1

Dataset Directory Structure

├───Data
│   ├───Test
│   │   ├───benign # 500
│   │   └───malignant # 500
│   ├───Train
│   │   ├───benign # 3500
│   │   └───malignant # 3500
│   └───Validate
│       ├───benign # 1000
│       └───malignant # 1000
│           └───.ipynb_checkpoints

Requirements

• Python 3.10.11
• TensorFlow 2.11.0
• NumPy
• Matplotlib
• Anaconda Navigator
• Your Skin Cancer Image Dataset (not provided, you can use your own dataset)

Installation

1. Install Anaconda Navigator:

Click here to install Anaconda Navigator.

2. Install Tensorflow :

Follow the steps in this video to install Tensorflow.

2.1 Create a Conda Environment:

  $ conda create -n py310 python=3.10

2.2 Activate Conda Environment:

  $ conda activate py310

2.3 Install CUDA toolkit & CUDNN Library:

  $ conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0

2.4 Install Tensorflow:

  $ python3 -m pip install "tensorflow<2.11"

2.5 Test GPU

  $ Python
  >> import tensorflow as tf
  >> tf.config.list_physical_devices('GPU')

Usage

Clone this Repository:

   git clone  https://github.com/Vipsy-123/Skin-Cancer-Classification-using-CNN.git

Dataset:

Download the Benign and Malignant images dataset from ISIC Archive

CNN Usage :

1. Open cnn1.ipynb
2. Run the cells to perform predictions on new skin cancer images

Transfer Learning Usage :

1. Open DensNet201.ipynb
2. You need to make 3 changes
   • Import your pre-trained model in Section 1. Installing Dependencies
   • Change input size in Section 2. Data Preparation & Augmentation
   • Change to your Model name in Section 3. Building the Model
     • e.g base_model = VGG16(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

Using models later :

You can see pre-trained models in models Directory to save Model Change the Model name in Section 4. Training the Model