Towards Non-Invasive, In-Office Detection of Early-Stage Bladder Cancer with Digitally Stained White-Light-Cystoscopy-Guided Polarization-Sensitive Optical Coherence Tomography
Chang, Shuang
0000-0002-3313-0683
:
2024-03-22
Abstract
Bladder cancer (BC) has a high recurrence rate (>50%) and requires frequent surveillance. However, the standard-of-care tool for BC surveillance, white light cystoscopy (WLC), has low sensitivity and specificity for detecting flat tumors like carcinoma in situ (CIS), a high-grade tumor with poor prognosis. Blue light cystoscopy (BLC) has shown promise but is limited by its accessibility, imaging artifacts, and low specificity in differentiating CIS against inflammations. To overcome the limitations of BLC, this dissertation aims to change how BC surveillance is performed by introducing a set of multimodal imaging tools that improves the sensitivity and specificity for in-office translation. The primary objective of this dissertation is to improve the detection of CIS through the development of digitally-stained-WLC-guided polarization-sensitive optical coherence tomography (PS-OCT) imaging.
First, digital staining was performed on WLC images to produce BLC-like images with fluorescence on suspicious lesions. The resulting digitally generated BLC (dgBLC) is a wide-field imaging tool with high sensitivity for CIS detection. Next, to remove the common environmental imaging artifacts in BLC imaging, computation artifact removal methods were introduced. The resulting enhanced BLC frames achieve better perceptual quality than original BLC frames and recover important diagnostic information in tissues. Lastly, to address the issue with low specificity of BLC and consequently our resulting dgBLC technology, PS-OCT was used, which is a variant of OCT with enriched sensitivity to collagen. With PS-OCT, quantitative information about tissue birefringence and optical attenuation coefficient were extracted. These metrics were shown to allow better differentiation of CIS from benign tissues (improve specificity of detection from 78.5% to 95%) and proved to have high correlations with morphological features computed from histology.
The proposed technology fills a critical gap in the standard of care for early detection of flat, high-grade carcinomas (i.e., CIS), especially in recurrent BC patients with significant inflammation. This dissertation research paves the way for a multimodal clinical tool with improved sensitivity and specificity in detecting CIS tumors for both in-office and surgical use.