NLM DIR Seminar Schedule

Abstract:

The technique of deep learning, or artificial intelligence (AI) broadly, has been employed in a lot of medical informatics research driven by imaging data. Topics such as classification and object detection have been actively studied in the field. It is well known that deep learning is data hungry technique. However, a higher data quantity doesn’t always guarantee higher performance. Data quality is also important for training of a robust deep learning model. In our studies using medical imaging data, quality factors include image sharpness, resolution, image labeling, specular reflection, data noise, etc. In this talk, several deep learning techniques will be introduced for dataset filtering, data augmentation, data enhancement, etc. These techniques are used to recode our cervical cancer datasets to achieve higher data quality.