Endoscopic Measurement of Lesion Size: an Unmet Clinical Need
Yinong Zhu,Wei Liu,Lifan Zhang,Bing Hu
DOI: https://doi.org/10.1097/cm9.0000000000002882
2023-01-01
Abstract:In recent years, the field of endoscopy has experienced significant technological advancements. These developments have facilitated the efficient use of endoscopy for precise examination and minimally invasive treatment of various lesions. A key parameter considered during both endoscopic diagnosis and treatment is the size of the lesion. In colon polypectomy, the size of the polyp is linked to the probability of malignancy and the likelihood of complications. Similarly, in the case of early gastric cancer, larger tumors are often associated with an increased likelihood of lymph node metastasis, and whether to perform endoscopic submucosal dissection (ESD) is partially dependent on the size of the lesion. Therefore, accurate measurement of lesion size may significantly impact the diagnosis and treatment plan for the patient. However, at present, there is no standardized method for quantitative measurement during endoscopy. This lack of a universal technique hinders the ability of clinical guidelines to provide precise recommendations for treatment. Currently, lesion size is being estimated using rudimentary tools such as premeasured endoscopic forceps as a reference. In 1995, Vakil[1] published a review highlighting the unreliability of current clinical methods for lesion size estimation during endoscopic examinations due to technical limitations. The use of image processing and stereoscopic techniques was proposed in Vakil's research as potential solutions for accurate measurement. In recent years, with advancements in medical and information technology, several methods involving optical component improvement, computer vision, and machine learning techniques have been proposed for endoscopic measurement. Despite these developments, related clinical trials are still rare, and a standardized, reliable, and valid measurement model has yet to be established. We categorized the existing endoscopic measurement techniques into three distinct groups and provided a concise overview of the experimental results obtained from relevant research studies. The strategy for literature search and the criteria for selecting eligible studies are illustrated in Supplementary Figure 1, https://links.lww.com/CM9/B856. Furthermore, we conducted an assessment of the advantages and limitations associated with each approach, while also delving into possible methodological improvements that could enhance the scope of future clinical investigations. Visual Estimation Method: The visual estimation method involved utilizing a reference object of known size placed around a lesion to provide an estimate of its size without the need for software programing. It is one of the simplest measurement techniques, and it is therefore the most widely used method in clinical applications. However, relying solely on manual observation inevitably increased the likelihood of errors. In addition, since the reference object (e.g., biopsy forceps[2]) must be placed as close as possible to the target lesion during estimation, it greatly increases the probability of secondary damage or even the probability of penetration. The use of alternative objects, such as rubber discs,[3] may lead to erratic measurements and unnecessary prolongation of the examination. The intuitive comparison method offers a simple and cost-effective approach to size measurement. Experienced physicians can obtain relatively accurate estimates by such a method. However, its large margin of error renders it unsuitable for meeting the stringent requirements of size measurement in clinical practice. Therefore, the introduction of computer-aided measurement tools has become increasingly important. Artificial Intelligence (AI)-Assisted Measurement: The versatility of AI in endoscopy has increased due to significant advancements in computer science. AI is now utilized in various ways in the field of endoscopy, such as image analysis, automated diagnosis, quality control,[4] and lesion measurement. Two different studies have validated the effectiveness of different AI technologies in endoscopic measurement, with both AI systems demonstrating commendable performance. Ali et al[5] developed a method for the automatic quantification of Barrett's esophagus. Sophisticated deep learning techniques were employed to predict the distance between an endoscopic camera and gastric folds. In parallel, a separate real-time AI system was developed to automatically identify and quantify Barrett's esophagus by recognizing distinctive landmarks at the margin of the lesion. The integration of the resulting depth maps and quantified Barrett's esophagus figures enabled the projection of endoscopy images onto three-dimensional (3D) space. The ground truth was obtained from a 3D-printed model reconstructed from the endoscopy video and measured by Vernier calipers and grid paper. Kwak et al[6] developed another AI to assess the size of colon polyps. Researchers utilized the W-Net model for vessel segmentation, employing retinal image datasets and colonoscopy images, to develop the bifurcation-to-bifurcation distance measuring method. This method was subsequently tested on endoscopic images, and the reference standard of the polyp size was measured by a ruler before fixation. The comprehensive results of two studies on AI techniques are presented in Supplementary Table 1, https://links.lww.com/CM9/B856. Although they provided accurate measurements, these studies are both limited by their small sample sizes and narrow focus on specific lesion types. In addition, there are concerns about the transparency and interpretability of AI itself. AI relies on machine learning algorithms, processing data in ways that are not easily understood or audited by humans, which naturally leads to systematic skepticism toward it.[7] Additional studies on the interpretability of AI are necessary, not only in the context of clinical trials but across all fields, to fully harness the potential of this technology. Despite these limitations, AI-assisted measurement has several advantages. It is highly accurate and does not require additional equipment to obtain measurements. This makes it a valuable tool for advancing multicenter trials and even truly extending its application to clinical settings. Optical-Assisted Measurement: With the advancement of optical manufacturing, it has become feasible to incorporate optical instruments to aid endoscopic measurement. An endoscope is a type of monocular camera system. In such a system, size estimation can be accomplished by obtaining the pixel size of a lesion since there is a specific function between pixel size and actual size, which solely depends on the distance from the endoscope lens to the lesion. This function can be determined in an in vitro environment. Based on this principle of a monocular camera system, during the endoscopy procedure, if the distance between the tip of the endoscope and lesion can be obtained in real time, it would subsequently allow for size measurement [Supplementary Figure 2, https://links.lww.com/CM9/B856]. Two studies were conducted based on this principle, and their detailed results are presented in Supplementary Table 2, https://links.lww.com/CM9/B856. Oka et al[8] designed an endoscopic system generating a grid scale to measure the lesion size. The width of the grid scale was constantly adjusted according to the distance between the tip of the endoscope and lesion, which was calculated using the amount of laser light reflected from the lesion through an optical probe inserted into the instrument channel [Supplementary Figure 3, https://links.lww.com/CM9/B856]. Endoscopists can thus estimate the size by comparing it with the grid scale. Yoshioka et al[9] developed a virtual scale function for endoscopic measurement. A red laser beam was emitted diagonally from the tip of the endoscope, and its position on the object changed according to the movement of the endoscope. When the image sensor detects the position of the laser spot, it computes the distance between the endoscope and the object illuminated by the laser [Supplementary Figure 4, https://links.lww.com/CM9/B856]. Subsequently, the length of the virtual scale was adaptively modified based on this calculated distance, enabling endoscopists to estimate the size of the object using the scale. In addition to two-dimensional (2D) size estimation, an alternative method exists for obtaining a 3D reconstruction of the lesion. Furukawa et al[10] incorporated a micropattern projector to the tip of the endoscope to facilitate 3D reconstruction. The methodology involved projecting a wave grid pattern onto the object and capturing the image by endoscopic camera [Supplementary Figure 5, https://links.lww.com/CM9/B856]. The wave grid pattern underwent corresponding deformation with the 3D shape of the projection plane, and the distorted pattern was extracted from the image for the calculation of reconstruction. Subsequently, depth values for all pixels were estimated to produce a dense 3D shape reconstruction. However, optics-assisted measurement has intrinsic shortcomings. Optical components necessitate additional manufacturing processes, which considerably impede their widespread adoption and may potentially compromise experimental reproducibility. The incorporation of such additional attachments to an endoscope may also constrain the operation in actual clinical conditions since it might enlarge the size of the endoscope or have some extra requirements for the angle. Should optimization of optical component manufacturing be realized, such as miniaturizing the size of the attachment or amalgamating it in batches at the forefront of the lens, extensive sample experiments and even clinical applications for such research can be anticipated. Given the significance of lesion dimensions in the diagnosis and treatment of gastrointestinal disorders, an increasing number of studies have turned their attention toward devising a method capable of measuring these dimensions. An important issue in this field that warrants further exploration is the establishment of a "gold standard" for size measurement. In previous studies, various methods have been employed to determine the ground truth for size measurement. These include direct measurement of freshly excised tissue prior to fixation, experimentation with in vitro models of known dimensions, and direct measurement of a 3D-printed model of the digestive tract. However, mechanical excision can result in tissue deformation, potentially leading to discrepancies between post excision and in vivo measurements. The same dimensional changes also occur before and after specimen fixation, which should also be considered if using pathology results as the ground truth. Utilizing a reconstructed model introduces an additional system as a reference point, the reliability of which must be verified through experimentation. Employing a polyp model of known dimensions circumvents issues related to ground truth selection but fails to replicate the actual operating environment, resulting in significant deviations between experimental results and clinical outcomes. Thus, future experimental designs must address the question of how to balance the proportion of in vivo and in vitro experiments in research in this field, as well as how to best obtain ground truth values. The application of AI endoscopy is increasingly pervasive. Its reliance on computational operations, without additional equipment requirements, makes it highly amenable to widespread adoption. Researchers have also shown interest in the functionality of AI technology for diagnosing early-stage cancer and delineating its pathological boundaries, with several publications demonstrating high accuracy. The integration of AI capabilities for comprehensive early-stage cancer diagnosis, delineation of pathological boundaries, and measurement of delineated area sizes would enable the practical implementation of endoscopic guidelines, enabling physicians to make more precise assessments of patient treatment and allowing a greater number of patients to benefit from minimally invasive endoscopic interventions. Besides, endoscopy application is beyond the confines of the digestive tract. In the field of the respiratory system, bronchoscopy has demonstrated precise localization capabilities through the utilization of electromagnetic navigation.[11] The integration of electromagnetic navigation technology into gastrointestinal endoscopy, in combination with the monocular camera size measurement principle, presents a theoretical possibility for estimating lesion size. In contrast to the obfuscated decision-making methodology inherent in AI technology itself, optics-assisted measurement provides more immediate estimations based on established physical and mathematical principles. Consequently, the utilization of magnetic navigation technology ensures superior accuracy, resulting in highly precise determination of lesion size. This approach may establish a highly reliable and trustworthy system, providing new possibilities for lesion size measurement in endoscopy. In conclusion, the implementation of lesion size measurement under endoscopy can benefit many patients. For the purpose of realizing and optimizing endoscopic measurement, future research should focus on developing more sophisticated algorithms capable of accurately measuring lesion size, irrespective of variations in endoscopic equipment and image quality. Additionally, efforts should be made to establish a scalable and repeatable measurement model that can be applied across different endoscopic platforms and settings to truly integrate it into clinical practice. Funding This work was supported by grants from Chengdu Science and Technology Project (No. 2022-YF05–01722-SN), "1·3·5" Project for Disciplines of Excellence, West China Hospital, Sichuan University (No. ZYJC21011), and Sichuan University Basic Research Guiding Special Project of Engineering Research Center of Ministry of Education (No. SCU2023D015). Conflict of Interest None.