Skip to content

Latest commit

 

History

History
56 lines (44 loc) · 3.08 KB

My_project.md

File metadata and controls

56 lines (44 loc) · 3.08 KB
Raw

Using Convolutional Neural Network to classify pictures of pathology of colorectal cancer

(Outline)

[TOC]

Goal

I would like to build a my own CNN (Convolutional Neural Network) and use this CNN to classify two kinds of colorectal cancer.

Applications

Firstly, it can help clinical practitioners to classify the two kinds of colorectal cancer more accurately. Secondly, I can learn more from the practice of image marchine learning.

Expected outcomes

The accuracy rate of the validation dataset and training data dataset are expected up to 85% based on the limited small sample size.

Data sources

These pictures have been labeled by pathologists at the University Hospitals Coventry and Warwickshire link, which included two classifications: benign (74) and malignant (91). All 165 pictures were split into two datasets: training dataset (85) and test dataset (80). The composition of the dataset is as follows.

Split Warwick-QU
Training benign : 37
malignant : 48
Test benign : 37
malignant : 43

Tabel of classification of colorectal cancer
Tabel of classification of colorectal cancer

Sample of picture
Sample of picture

Data processing

  • Transform all pictures to "tfrecord" file so that Colab can read these data. (xml-- csv --tfrecord , train_images, train_labels, test)
  • I will learn how to load my own dataset into Colab link.

(Delete) Transfer learning with tensorflow hub according to this website.

  • Download the classifier
  • Load my dataset
  • Run the classifier on a batch of images
  • Download the headless model
  • Attach a classification head
  • Train the model
  • Check the predictions and evaluate the model
  • Compare the effect of different classifier, like mobilenet, inception v3,,,
  • Export the model

Build a convolutional neural network according to below websites and link. (if possible)

Only use CNN to train because of limited time and incompatible equipments

(Delete) Compare the effect of convolutional neural network, and neural network. (if possible)

Questions:

  • I am not sure whether I can finish the above analysis by using only Colab. If not, I may need set up a enivironment the transfer learning.
  • My person computer is competent or not.
  • How to prepare a well dataset for Colab.
  • The training smample size my be not enough.

(have been solved)