Transfer Deep Learning of Medical Images: Mammography classification

Mammography is a type of medical imaging used to diagnose and screen for breast cancer. High misdiagnosis rates cause many patients to go untreated and others to go through unnecessary, distressing and sometimes harmful treatment.
This project aims to create a system that would automatically diagnose a patient based on given mammography scans, diminishing the effect of human error and freeing up physician time for deeper examination of edge cases.
To create the classification system, the Digital Database for Screening Mammography was used to train a deep convolutional neural network called Inception. The method used to train the model was an improvement upon a method called Transfer Learning. Additionally, a method called TI Pooling was implemented to further increase the model’s ability to correctly classify the scans.
The result of the system: A 4.2% increase in classification accuracy and a 0.045 increase in AUC score when compared to the basic transfer learning model. A 15.1% increase in classification accuracy compared to the statistical physician diagnosis. The large gain in the accuracy was achieved indicates that this is a promising field of research that could have great benefits if used in real world applications.

Transfer Deep Learning of Medical Images: Mammography classification