Intelligent Image Feature Matching for Small Intestine Capsule Endoscopy

描述（由申请人提供）：小肠疾病影响近1900万美国人。通过最近引入的“胶囊内窥镜检查”，这些患者的护理得到了显着改善。这需要吞下一个药丸形状的成像装置，该装置将图像无线传输到外部接收器。这首次使小肠粘膜的非侵入性视觉成像成为可能。该技术有可能为30 - 50%的患者提供诊断，即使在进行了彻底的食管胃十二指肠镜检查（EGD）和结肠镜检查后，隐匿性胃肠道出血仍无法解释。然而，胶囊内窥镜检查需要人工检查设备拍摄的大约50,000张图像。此外，由于囊的运动性未知，在图像中看到的特定病变的定位具有挑战性。这促使我们的小组研究使用视觉运动跟踪来加强胶囊内窥镜研究的回顾。我们建议从单个胶囊图像中创建小肠的单一重建图像表示。这将提供三个关键优势：1)图像将合并，以消除对同一肠腔区域的重复检查，从而减少总检查时间。2)通过建立图像间的空间关系，实现病灶定位；3)如果图像不重叠，我们将能够识别未成像的区域，这是目前的胶囊内窥镜系统无法做到的。我们的第一阶段工作包括开发智能图像特征匹配软件，并使用胶囊内窥镜平台的图像进行评估。我们将评估1)特征识别，2)图像之间的特征匹配，以及3)基于胶囊内窥镜捕获的图像测量腔内距离的能力。如果成功，我们设想第二阶段的努力，将利用我们的智能特征匹配技术来构建一个先进的GI图像浏览系统。我们建议从单个胶囊图像中创建一个小肠的马赛克表示。该系统将在试点人体研究中进行评估，以调查其快速可视化和定位潜在病变的能力。我们的长期目标是为胃肠病学家提供一种非侵入性的小肠可视化工具，这种工具可以快速解释并准确定位病变。胶囊内窥镜首次实现了小肠远端粘膜的无创视觉成像。然而，耗时的人工检查和无法定位病变阻碍了它的发展。我们提出的软件技术将为胃肠病学家提供一个快速有效的工具来审查胶囊内窥镜检查数据。

DESCRIPTION (provided by applicant): Diseases of the small intestine affect nearly 19 million Americans. The care of these patients has improved significantly via the recent introduction of "capsule endoscopy". This involves swallowing a pill-shaped imaging device which wirelessly transmits images to an external receiver. For the first time, this has enabled non-invasive visual imaging of the small intestinal mucosa. The technology has the potential to provide a diagnosis to the 30 - 50 % of patients whose occult gastrointestinal bleeding remains unexplained even after thorough workup with esophagogastroduodenoscopy (EGD) and colonoscopy. However, capsule endoscopy involves manual review of approximately 50,000 images taken by the device. Furthermore, localization of a particular lesion seen in the images is challenging due to the unknown motility of the capsule. This prompted our group to investigate the use of visual motion tracking to enhance review of capsule endoscopy studies. We propose to create a single reconstructed image representation of the small intestine from the individual capsule images. This will provide three key advantages: 1) Images will be combined to eliminate redundant review of the same area of the intestinal lumen, thereby decreasing total review time. 2) By creating spatial relations among images, lesion localization will be possible; and 3) If images are non-overlapping, we will be able to identify unimaged areas, which current capsule endoscopy systems cannot do. Our Phase I effort involves development of the intelligent image feature matching software and evaluation using images from a capsule endoscopy platform. We will evaluate 1) feature identification, 2) feature matching between images, and 3) the ability to measure intraluminal distances based on the captured images of the capsule endoscope. If successful, we envision a Phase II effort that will harness our intelligent feature matching technology to construct an advanced GI image browsing system. We propose to create a single mosaiced representation of the small bowel from the individual capsule images. The system will be evaluated in a pilot human study to investigate its ability to quickly visualize and localize potential lesions. It is our long term goal to provide gastroenterologists with a non-invasive tool for visualization of the small intestine which is quick to interpret and allows accurate localization of lesions. Capsule endoscopy for the first time has enabled non-invasive visual imaging of the distal small intestinal mucosa. However, it is hindered by time-consuming manual reviews and an inability to localize lesions. Our proposed software technology will provide gastroenterologists with a quick and efficient tool to review capsule endoscopy data.