Robin Elizabeth Yancey

Robin Elizabeth Yancey

Each woman living in the United States has about 1 in 8 chance of developing invasive breast cancer. The mitotic cell count is one of the most common tests to assess the aggressiveness or grade of breast cancer. In this prognosis, histopathology images must be examined by a pathologist using high-resolution microscopes to count the cells. Unfortunately, this can be an exhaustive task with poor reproducibility, especially for non-experts. Deep learning networks have recently been adapted to medical applications which are able to automatically localize these regions of interest. However, these region-based networks lack the ability to take advantage of the segmentation features produced by a full image CNN which are often used as a sole method of detection. Therefore, the proposed method leverages Faster RCNN for object detection while fusing segmentation features generated by a UNet with RGB image features to achieve an F-score of 0.508 on the MITOS-ATYPIA 2014 mitosis counting challenge dataset, outperforming state-of-the-art methods.

Robin Elizabeth Yancey

With technological advances leading to an increase in mechanisms for image tampering, fraud detection methods must continue to be upgraded to match their sophistication. One problem with current methods is that they require prior knowledge of the method of forgery in order to determine which features to extract from the image to localize the region of interest. When a machine learning algorithm is used to learn different types of tampering from a large set of various image types, with a large enough database we can easily classify which images are tampered. However, we still are left with the question of which features to train on, and how to localize the manipulation. In this work, deep learning for object detection is adapted to tampering detection to solve these two problems, while fusing features from multiple classic techniques for improved accuracy. A Multi-stream version of the Faster RCNN network will be employed with the second stream having an input of the element-wise sum of the ELA and BAG error maps to provide even higher accuracy than a single stream alone.