The goal of the individual assignment is to use deep learning, in particular unsupervised Convolutional Neural Network(CNN), to make a tool for estimating pelvic floor problems. This is done according to the following research question:
Can we use unsupervised deep learning to obtain clinical relevant information from pelvic floor ultrasound images?
Unsupervised learning is used to create a structure of the unlabeled data. The image is going to be transformed into a vector using different layers of the CNN. This vector like structure is analyzed using principle component analysis and/or K-means clustering to find any relation in deeper layers of this network. These layers are examined to find correlation between the different images. The available data set consists of ultrasound images from pregnant woman from 12, 36 weeks and 6 months after delivery. By using these techniques, the images and data of 258 women is combined to create a prediction for possible changes in functionality of the pelvic floor during/after pregnancy.
The output data is going to be analyzed where the goal is to classify a certain group of high risk patients.