Jiawei Zhong, Hongliang Ren, Qin Chen & Hui Zhang
Modern endoscopy technology has revolutionized minimally invasive surgical procedures across various medical applications, including gastrointestinal (GI) tract screening, bronchi examination and transsphenoidal surgery. To address the inherent challenges in these applications, it is crucial to achieve accurate localization, mapping and 3D reconstruction within the in-vivo cavity environment. Deep learning-based techniques such as Simultaneous Localization and Mapping (SLAM), Structure from Motion (SfM) and Neural Rendering offer promising solutions to this complex problem. In this literature review, we provide an overview of 42 previous research papers that have addressed the challenges of localization, mapping and reconstruction using different deep learning methods. We introduce a novel taxonomy to categorize these previous works and briefly discuss their respective advantages and disadvantages. Furthermore, we present a comprehensive evaluation methodology to quantitatively compare the most representative research efforts in this field. Finally, based on our findings, we summarize the current challenges and future opportunities in the realm of endoscopy localization, mapping and 3D reconstruction