A Novel Hybrid Endoscopic Dataset for Evaluating Machine Learning-based Photometric Image Enhancement Models

Abstract

Endoscopy is the most widely used medical technique for cancer and polyp detection inside hollow organs. However, images acquired by an endoscope are frequently affected by illumination artefacts due to the enlightenment source orientation. There exist two major issues when the endoscope's light source pose suddenly changes: overexposed and underexposed tissue areas are produced. These two scenarios can result in misdiagnosis due to the lack of information in the affected zones or hamper the performance of various computer vision methods (e.g., SLAM, structure from motion, optical flow) used during the non invasive examination. The aim of this work is two-fold: i) to introduce a new synthetically generated data-set generated by a generative adversarial techniques and ii) and to explore both shallow based and deep learning-based image-enhancement methods in overexposed and underexposed lighting conditions. Best quantitative results (i.e., metric based results), were obtained by the deep-learnnig -based LMSPEC method, besides a running time around 7.6 fps.

Multi-Scale Structural-aware Exposure Correction for Endoscopic Imaging

“…In a prospective study, Garcia-Vega et al [5] compared various image enhancement methods, using a recent dataset containing synthetically generated over-and underexposed endoscopic frames which are paired with their noncorrupted ground truth image. The authors assessed the capabilities of different IE methods to enhance the quality of endoscopic images, while maintaining a high degree of fidelity.…”

“…The dataset used in this contribution is a combination of three different existing datasets (EAD [11], EDD [12] and HyperKVisir [13]) using the procedure described in [5]. The authors used image-to-image translation to take unmodified endoscopic frames and generated frames with over-and underexposure artefacts.…”

“…García-Vega et al proposed a paired "normal-exposed" image dataset [5,6] to assess the ability of machine learning (ML) based methods to correct the effects of non-optimal lighting conditions. However, the need for both accurate and real-time IE techniques highlighted the shortcomings of most current methods.…”