Window Level Optimization for Pectoral Muscle Segmentation

Saksham Jain      Wenshao Zhu

saksham.jain@duke.edu     wenshao.zhu@duke.edu

Paper PDF

Digital Imaging and Communications in Medicine (DICOM) is a standard used for storing data obtained using medical imaging and is key to interoperability between medical devices. The data stored in DICOM images needs to be processed for use such as converting image values into Hounsfield units for CT imaging. This data can then be further processed using window levels which emphasizes values within a certain range. By setting a window, certain tissue and pathologies can be better identified making the technique useful for identifying key information. We were interested in seeing whether optimization via a CNN could optimize this setting not only for use in classification and segmentation tasks but also in creating better windows for human use. We looked to optimize the window level for the task of pectoral muscle segmentation and found that optimizing the window level improved the ability of our UNet to segment the input images however our current results do not suggest the benefit of this windowing extends to human use.

Paper: Paper PDF