Specular Reflection Removal on Smart Colposcopy Images using Deep Learning Inpainting Model for Enhanced Grading of Cervical Cancer

Abstract

Cervical cancer is a significant global health concern, ranking as the fourth most newlinecommon cancer among women, primarily caused by Human Papillomavirus (HPV) newlineaffecting the lower uterus. Despite preventive measures like HPV vaccination and newlinescreening programs, many women hesitate due to invasiveness. Smart colposcopy, an newlineadvanced non-invasive approach, captures cervix images for examination. However, white newlinespecular reflections caused by body moisture pose challenges, hindering accurate analysis newlineand potentially leading to misclassification of dysplasia regions. This research aims to newlineimprove cervical cancer grading by identifying and removing specular reflection from newlinesmart colposcopy images. Initial focus lies on specular reflection identification, employing newlineRGB and XYZ color spaces for optimal detection. The proposed intensity-based threshold newlinemethod accurately identifies specular reflection on XYZ color, overcoming challenges newlineposed by vaginal discharge and acetowhite regions. newlineIn the second phase, pixel-wise segmentation models like Fully Convolutional newlineNeural Network (FCN), SegNet, and UNet Model are employed. On comparison analysis newlineof the segmentation model, the UNet model indeed demonstrates higher accuracy. newlineHowever, when it comes to the intersection of Union, the UNet model falls short due to newlinethe overlapping of segmentation. To address this limitation, different versions of the UNet newlinemodel are compared, and the UNet++ model emerges as the most promising, exhibiting newlinehigher intersection of union metrics. Subsequently, the UNet++ model is fine-tuned to newlineoptimize its performance in segmenting reflection regions. After segmentation of the newlinereflection, the empty region should be filled with neighboring pixels to improve the newlinequality of the images. A novel Bilateral-based Convolutional Inpainting model fills newlineeliminated regions, improving image quality. This model outperforms traditional methods, newlineparticularly in medical image applications, showcasing efficacy across different masking newlineratios.