Automated layer segmentation of bladder OCT images for enhanced detection of transitional cell carcinoma

Optical coherence tomography(OCT) imaging of bladder is gaining recognization due to the capability of noninvasive cross-sectional imaging of the bladder at the micron-level resolution and a relatively large field of view. Previous studies have shown the potential of OCT image to enhance detection of bladder transitional cell carcinoma(TCC). However, quantitative OCT image analysis for affirmative identification of bladder tumor remains a challenge[1]. Here, we report a novel method to enhance detection of TCC based on OCT images by analyzing anatomical and textural alteration of bladder. Specifically, OCT images are first processed with Dual Tree Complex Wavelet Transform denoising algorithms to reduce image speckle noise. Then, the layer segmentation method that mainly based on a dual path graph searching algorithm is performed on the denoised images to delineate three layers of bladder. The segmentation results show improved effectiveness and robustness in comparison to conventional graph theory based method. With layer segmentation, multiple measurements including layer thickness and texture can be quantified. The significant difference in quantified metrics between TCC and normal bladder indicate the potential use of those metrics for TCC identification. The proposed method provides valuable insights into TCC and has the potential to enhance the detection of tumor in the clinic.