Computer-Aided Diagnosis With Monochromatic Light Endoscopy for Scoring Histologic Remission
Computer-Aided Diagnosis With Monochromatic Light Endoscopy for Scoring Histologic Remission in Ulcerative Colitis
The evaluation of endoscopic and histologic remission is the cornerstone in drug development and treatment planning in patients with ulcerative colitis (UC). Treatment targets should be reproducible, easy to assess, and predictive of the further disease course. At this time, the evaluation of remission continues to depend on a human interpretation of the degree of inflammation and for this it is prone to variability. 1 Computer-aided diagnosis (CAD) can improve the consistency of this evaluation. Histology is currently the best predictor of sustained clinical remission in UC. 2 In the outer extremes of the remission spectrum, endoscopy has difficulties to delineate the subtlety of the degree of remission and is therefore less valuable to predict sustained clinical remission. 3 In active UC, the mean intercryptal distance and wall thickness of the crypts increases. Although not incorporated in the current histologic scores, changes in the mucosal pericryptal vasculature are associated with the degree of inflammation. 4 , 5 These microscopic vascular changes can currently only detected in vivo by confocal laser endomicroscopy. 5
Description of Technology
We describe a new CAD technique using images from a prototype endoscope with a single short wave-length monochromatic LED light illumination (Fujifilm, Tokyo, Japan). This modality enables the real-time evaluation of the mucosal architecture (including crypts, pericryptal capillaries, and bleeding) up to a depth of approximately 50–200 μm ( Figure 1 ). This technique does not require additional administration of intravenous contrast agents. To optimize optical focus, a distal attachment cap is used to maintain an optimal distance from the colonic mucosa.
Open full size image Figure 1Details of the short wave length monochromatic light image showing the colonic mucosal structures up to a depth of 50–200 μm.
We hypothesized that histologic nonremission (a Geboes score of ≥2B.1: neutrophil infiltration in the lamina propria) is associated with mucosal capillary congestion, capillary leakage, and bleeding. These changes are accumulating with an increasing degree of inflammation. The in vivo assessment of superficial capillary vascular structures could predict histologic activity in UC. In this study, we prospectively assessed patients with UC for planned endoscopy at the outpatient clinic of the University Hospitals of Leuven (Belgium) (ethical approval S59405). We applied no limits for age, disease duration, or treatment to assess the complete spectrum of the UC phenotype. Patients with unspecified inflammatory bowel disease, Crohn’s colitis, ostomy, total colectomy, and ileal pouch anal anastomosis, as well as pregnant women, were excluded.
Fifty-eight patients (53% male; median age, 41 years; interquartile range, 38–56 years; disease duration 7.1 years; interquartile range, 2.4–16.4 years) with 113 evaluable segments (89% rectum or sigmoid) were included. White light and monochromatic light images were obtained in different magnification modes and biopsies were sampled in the corresponding region. All images were scored for the endoscopic subscore of the endoscopic subscore of the Mayo score (MES) and the Ulcerative Colitis Endoscopic Index of Severity (UCEIS). The biopsies were scored according to the Geboes score.
Based on these data, the CAD algorithm was built in several steps. In the first step, we evaluated the images for the presence of mucosal or luminal bleeding related to capillary leakage. This evaluation was made based on an automated feature extraction technique. This provided a value per image indicating the number of pixels with bleeding (NPBL). The second step was extraction of the vascular pattern from the monochromatic light images with a morphologic hessian-based vessel recognition technique. Based on the extracted vascular map, we calculated the density of the mucosal vessels per pixel. From this density map, we could calculate the number of pixels with high density (NPHD). Finally, we combined the 2 CAD steps into 1 algorithm to optimize the correlation with the histologic score ( Figure 2 ).
Open full size image Figure 2Schematic presentation of the computer-aided algorithm to assess histologic remission in ulcerative colitis based on endoscopic images.
The algorithm first takes the NPBL assessment into account. If the NPBL is high, then the images are classified as nonremission. If the NPBL is low, then the images are further evaluated for NPHD. Images with a high NPHD are classified as nonremission. The current automated CAD algorithm detects histologic remission with a high performance (sensitivity of 0.79 and specificity of 0.90) compared with the UCEIS (sensitivity of 0.95 and specificity of 0.69) and MES (sensitivity of 0.98 and specificity of 0.61), resulting in a positive predictive value for histologic remission of 0.83, 0.65, and 0.59 for the CAD algorithm, UCEIS, and MES respectively. Receiver operating characteristics and correlation with histological scoring are shown in Supplementary Material . The CAD algorithm detects histologic remission with high accuracy (86%) versus the MES (74%) and the UCEIS (79%) and indicates in an objective way if a treatment target of histologic remission is reached.
Take Home Message
We present a novel, real-time, automated evaluation of specific changes in the mucosal pericryptal vascular structures associated with UC disease activity. These relevant changes are linked to the infiltration of inflammatory cells in the colonic wall, but are currently difficult to quantify in vivo. This endoscopic system is the first that optically visualizes the pericryptal vessels without any additional need for a contrast agent, by the application of an innovative illumination with a single wave length monochromatic light source. We developed a real-time CAD algorithm, based on these newly highlighted features that allows to assess histologic activity in UC with high accuracy. The automated system has the potential to provide instant and accurate histologic scoring of the colonic mucosa. Therefore, it is a promising tool for objective evaluation of endoscopic/histologic mucosal healing in a treat-to-target approach in patients with UC.
In its current state, the CAD algorithm does not replace standard biopsies to confirm remission and to assess other features (infections and dysplasia). Further validation in an independent cohort is necessary. Next, we will explore the sensitivity to change and the predictive value of the CAD algorithm. The concept of the automated evaluation of the mucosal capillary structures based on this technique provides potential other applications including automated colonic dysplasia detection.