A Similarity Measure of Histopathology Images by Deep Embeddings
This work addresses the problem of efficiently comparing histopathology images for medical professionals, but it is incremental as it builds on existing deep learning and similarity methods.
The study tackled the challenge of content-based comparison of high-resolution histopathology images by proposing a similarity measure using deep embeddings at multiple magnification levels, achieving a maximum accuracy of 93.18% for top-5 search at 5x magnification.
Histopathology digital scans are large-size images that contain valuable information at the pixel level. Content-based comparison of these images is a challenging task. This study proposes a content-based similarity measure for high-resolution gigapixel histopathology images. The proposed similarity measure is an expansion of cosine vector similarity to a matrix. Each image is divided into same-size patches with a meaningful amount of information (i.e., contained enough tissue). The similarity is measured by the extraction of patch-level deep embeddings of the last pooling layer of a pre-trained deep model at four different magnification levels, namely, 1x, 2.5x, 5x, and 10x magnifications. In addition, for faster measurement, embedding reduction is investigated. Finally, to assess the proposed method, an image search method is implemented. Results show that the similarity measure represents the slide labels with a maximum accuracy of 93.18\% for top-5 search at 5x magnification.