Classification of Computed Tomography Images with Pleural Effusion Disease Using Convolutional Neural Networks

On the present work we use two different convolutional neural nets architectures for the classification of computed tomography images with pleural effusion disease. We decided to use the convolutional neural networks due to the great advances achieved by this kind of nets in image classification problems. We work with a real-world data anonymized and provided by an Imagenology Department of public hospital from Chile. The data was classified by medics of the hospital. Due to the limitations on graphics resource, we decided training the algorithms from scratch, avoiding overfitting with regularization techniques and optimizing the training process programming callbacks. For testing, we used a set of 1,000 images and evaluate with classification metrics like True positive rate, True negative rate and Accuracy. Results achieved were not optimal due to overfitting of algorithms. For future works, we will use other architectures of convolutional neural networks and with Transfer learning technique on the architectures.