New SOTA on Malaria Detection

The motivation of this project comes from a PyImageSearch tutorial Deep Learning and Medical Image Analysis with Keras. In that tutorial, Adrian Rosebrock of PyImageSearch briefed about medical tests condicted for testing malaria and how he was able to achieve SOTA score over the work as discussed in Pre-trained convolutional neural networks as feature extractors toward improved parasite detection in thin blood smear images by Rajaraman et al. Adrian’s model was able to yield an accuracy score of 97% with a training time of about 54 minutes on Titan X GPU, whereas the model discussed in the paper took almost a day to train and generated an accuracy score of 95.9%.

So, I decided to challenge myself to see if I could apply the modern deep learning practices (as taught by Jeremy Howard in the course Practical Deep Learning for Coders v3) with the help of the fastai library. The good news is I did.

Here is the link to repo. https://github.com/sayakpaul/Malaria-Detection-with-Deep-Learning

4 Likes