University of Twente Student Theses

Login

The Effects of Active Learning on Computer-Aided Diagnosis in Multi-Disease Prediction of Chest X-rays

Janssen, Djordi (2019) The Effects of Active Learning on Computer-Aided Diagnosis in Multi-Disease Prediction of Chest X-rays.

[img] PDF
1MB
Abstract:The manual annotation of medical images in order to improve computer-aided diagnosis requires a profound level of expertise and is highly time-consuming. Active learning strategies have been widely used in cases where manual annotation is burdensome, as these strategies intend to optimize computers models while reducing the amount of necessary training data. In this research, four active learning strategies have been tested against a passive approach by using a convolutional neural network that has been trained on the National Institutes of Health Chest X-Ray dataset, containing 112,120 multi-label X-rays of the thorax. Additionally, we studied whether the performance of the learning strategies was affected by the amount of (available) training data and examined the impact of the strategies on the individual diseases. We found that active learning had a positive effect on both the area under the ROC curve and the ranking metrics when compared to passive learning, and that the optimal strategy was dependent on the size of the training data. The standard deviations from the average AUC of the individual diseases were higher when using active learning compared to passive learning. In conclusion, this study demonstrated that active learning is beneficial for training computer-aided diagnosis models on the prediction of multiple disease of the thorax.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:44 medicine, 54 computer science
Programme:Business & IT BSc (56066)
Link to this item:https://purl.utwente.nl/essays/79151
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page