Deep learning-based computer aided diagnosis (CAD) has been proposed to detect and classify prostate cancer lesions in multi-parametric Magnetic Resonance Imaging (mp-MRI) images. CAD requires their input images meet certain quality standards. In this work, we proposed a ResNet50-based model to filter out images not suitable as the input to the following lesion detection network. Taking unqualified images as positive cases, we obtained an area under ROC curve (AUC) of 0.8526 in test cohort, which helped to improve the performance of detection model and increased the interpretability by rejecting unqualified images with a reason instead of giving wrong results.
This abstract and the presentation materials are available to members only; a login is required.