Deep Learning with minimum coding and free hardware

The current work is proposing a very simple and fast python script to classify polyps in
colonoscopy images using Fastai deep learning. This work is the optimized version of the
previous tool: GitHub repository CNN4Polyps about colonoscopy polyp detection
(classification + localization into an image) with convolutional neural network - CNNs
(https://github.com/muntisa/Colonoscopy-polyps-detection-with-CNNs) using Keras. I
demonstrated that simple CNNs or VGG16 transfer learning could be used with a GPU to
create good classifiers able to detect a polyp in colonoscopy images.
Jeremy Howard and the team are proposing a faster python package for deep learning models
as a free course at Fast.ai (https://course.fast.ai/). Google Colab free virtual computers
(https://colab.research.google.com) with GPU support was used to run all the calculations.
The current dataset was generated with the previous project CNN4Polyps starting from a
public dataset: 910 images for training and 302 images for validation. All the models were
saved in the project folder.
The current script (Figure 1) demonstrated the ability to create a very accurate classifier for
medical imaging with an accuracy of 0.99 using resnet50 transfer learning fine tuning, only
5-10 lines of code, free GPU hardware (Nvidia K80), and free fastai package (based on
PyTorch library). The scripts to train the model or to make predictions are available at
https://github.com/muntisa/Fastai-Colon-Polyps.

Acknowledgements
I gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this
research (https://developer.nvidia.com/academic_gpu_seeding). In addition, the authors would like to
acknowledge support from the Galician Network for Colorectal Cancer Research (REGICC) (Ref. ED431D
2017/23). This work is also supported by “Collaborative Project in Genomic Data Integration (CICLOGEN)”
PI17/01826 funded by the Carlos III Health Institute from the Spanish National plan for Scientific and Technical
Research and Innovation 2013−2016 and the European Regional Development Funds (FEDER).