Member-only story
Using Transfer Learning to Detect Lung Infections
In the world of healthcare, it’s import to diagnose patiently correctly or else you can end up ingesting medicine that can end up making you feel worse with unwanted side effects. Luckily we have had groundbreaking results when incorporating artificial intelligence with other world problems. In this write up we’ll go over the use case of using computer vision and CNN’s to detect Pneumonia in X-ray images. Using X-rays are currently one of the best methods in identifying pneumonia, however, they do also have their limitations because they don’t show which kind of bacteria is attacking the lungs, so patients can still be misclassified with pneumonia or other viral infections such as acute bronchitis or other respiratory infections.
A popular research paper “CheXNet: Radiologist-Level Pneumonia Detection on Chest X-Rays with Deep Learning” ran studies where they evaluated the performance of classification for pneumonia and other infections while comparing their model scores with experienced radiologists. Using a pre-trained model (Imagenet) they were able to outperform radiologists using a 121-layer CNN.
Dataset:
We won’t be using the same images from the CheXNet study, alternatively, we’ll use data came from Kaggle. Special thanks to Radiological Society of North America (RSNA®) & US National Institutes of Health, The Society of Thoracic Radiology for the data.
