Skin Cancer Detection Tool README.md

Overview

The objective of this project is to build a Skin Cancer Detection Tool. The tool that we are creating is a segmentation model of spots (moles, melanomas, etc…) on microscopic images of the skin. To create this tool we will have to train a semantic segmentation AI model. The data that we use for that training is from The International Skin Imaging Collaboration.

File Descriptions:

  1. data.py: Contains functions to process and load the dataset, preprocess the images, and masks and to create TensorFlow datasets.

    • process_data(data_path, file_path): Reads the image and mask paths from the dataset.
    • load_data(path): Load training, validation, and test data.
    • read_image(x) and read_mask(x): Read the images and the masks respectively.
    • tf_dataset(x, y, batch=8): Create a TensorFlow dataset.
    • preprocess(x, y): Preprocess the images and masks.

  2. predict.py: Uses a pretrained model to make predictions on new images.

    • get_data(): Load test images from the INPUT_FOLDER.
    • Then, predictions are made using the loaded model and saved to the OUTPUT_FOLDER.

Setup & Requirements

Requirements:

  • python 3.x
  • pandas
  • numpy
  • scikit-learn
  • tensorflow 2.x
  • opencv-python

You can install these requirements using:

pip install pandas numpy scikit-learn tensorflow opencv-python

Steps to Run:

  1. Data Preparation:

    • Place your dataset in an appropriate directory.
    • Adjust the paths in the data.py script.
    • Run the data.py script to check if data is loaded properly.
      python data.py

  2. Predicting:

    • Place your test images in the INPUT_FOLDER.
    • Ensure the model path “segm_model” in predict.py corresponds to your trained model.
    • Run the predict.py script to make predictions.
      python predict.py

Notes

  • This tool currently segments the spots and saves the segmented images in the OUTPUT_FOLDER.
  • You might need to train the model first using your data to get the “segm_model”.
  • Ensure the directories mentioned in the scripts exist or are modified according to your directory structure.