high performance computing on graphics processing units: hgpu.org

Recently, a number of new image classification models have been developed to diversify the number of options available to prospective machine learning classifiers, such as Deep Learning. This is particularly important in the field of medical image classification as a misdiagnosis could have a severe impact on the patient. However, an assessment on the level to which a deep learning based Convolutional Neural Network can outperform a Support Vector Machine has not been discussed. In this project, the use of CNN and SVM classifiers is used on a dataset of approx. 55,000 images. This dataset was used to assess the classification potential of each methodology, in terms of training, implementation, and the ability to engineer parameters and features for successful classifications on a very large dataset. The use of CNN approaches is further broken down into the use of different frameworks, in this case Theano and Torch implementations. These are then compared to an SVM classifier by confusion matrix, training time and ease of use to assess which has the higher classification potential. Here it is seen that the Theano model outperforms the Torch model slightly for this task, by roughly 3% in the accuracy of the confusion matrix. The SVM meanwhile is shown to be very limited in its ability to classify such a large dataset. Furthermore, the SVM is shown to be limited in its ability to recognise the classes corresponding to the different levels of disease severity, achieving a classification accuracy of only 75% for the whole sample. These results suggest that the application of Deep Learning techniques currently have a very large advantage over SVM approaches in both accuracy and data handling, that to not natively avail of the computational power of Deep Learning.