Objective Methods of 5-Aminolevulinic Acid-Based Endoscopic Photodynamic Diagnosis Using Artificial Intelligence for Identification of Gastric Tumors

Abstract

Positive diagnoses of gastric tumors from photodynamic diagnosis (PDD) images after the administration of 5-aminolevulinic acid are subjectively identified by expert endoscopists. Objective methods of tumor identification are needed to reduce potential misidentifications. We developed two methods to identify gastric tumors from PDD images. Method one was applied to segmented regions in the PDD endoscopic image to determine the region in LAB color space to be attributed to tumors using a multi-layer neural network. Method two aimed to diagnose tumors and determine regions in the PDD endoscopic image attributed to tumors using the convoluted neural network method. The efficiencies of diagnosing tumors were 77.8% (7/9) and 93.3% (14/15) for method one and method two, respectively. The efficiencies of determining tumor region defined as the ratio of the area were 35.7% (0.0-78.0) and 48.5% (3.0-89.1) for method one and method two, respectively. False-positive rates defined as the ratio of the area were 0.3% (0.0-2.0) and 3.8% (0.0-17.4) for method one and method two, respectively. Objective methods of determining tumor region in 5-aminolevulinic acid-based endoscopic PDD were developed by identifying regions in LAB color space attributed to tumors or by applying a method of convoluted neural network.

Keywords: 5-aminolevulinic acid; LAB color space; neural network; photodynamic diagnosis.

Figure 1

Tumor lesions examined by LED-based system, (a1–c1) show lesion #19 (adenocarcinoma) (a1–c1) and lesion #20 (adenocarcinoma); (a1,a2) White light images where tumor locations are indicated by blue arrows; (b1,b2) photodynamic diagnosis images; (c1,c2) photodynamic diagnosis images with an annotated image overlaid, in which tumor region determined with reference to the pathology of the specimen is depicted in red and non-tumor region is depicted in green.

Figure 2

Tumor and non-tumor in the LAB space and the area assigned to tumor by method one: Points in the photodynamic diagnosis images are plotted with corresponding values in LAB color space, for tumor points (red circle) and non-tumor points (green cross). Points in LAB color space classified as tumors by method one are plotted as red-smoked areas.

Figure 3

Example of PDD images and the results applied by method one: (a1) Photodynamic diagnosis (PDD) image of lesion #19 (adenocarcinoma), (b1) PDD image with an annotated image overlaid, in which tumor region determined by method one is depicted in red; (a2) PDD image of lesion #20 (adenocarcinoma); (b2) The PDD image with an annotated image overlaid, in which tumor region determined by method one is depicted in red.

Figure 4

(a) Area efficiency (b) Area 1-specificity, when method one is applied.

Figure 5

Example of PDD images and the results applied by method two: (a1) Photodynamic diagnosis image of lesion #19 (adenocarcinoma, validated lesion), (b1) the result of method two overlaid to (a1), in which tumor regions determined by method two are depicted in brown and non-tumor regions are depicted in green; (a2) photodynamic diagnosis image of lesion #20 (adenocarcinoma, validated lesion),(b2) the result of the method two overlaid to (a2), in which tumor regions determined by the method two are depicted in brown and non-tumor regions are depicted in green.

Figure 6

(a) Area efficiency (b) Area 1-specificity for trained 12 lesions using method two.

Figure 7

(a) Area efficiency (b) Area 1-specificity for the validated 15 lesions using method two.

Figure 8

(a1,a2) Photodynamic diagnosis images of lesion #16 (validated lesion), which was detected as a tumor for the first time on photodynamic diagnosis endoscopy; (b1) annotated picture overlaid on (a1); (c1) the result of method two overlaid on (a1); (b2) annotated image overlaid on (a2); (c2) the result of method two overlaid on (a2); (d) white light image of the lesion in which the tumor location is indicated by a blue arrow; (e) blue light image of the lesion.

Figure 9

(a1–a4) Photodynamic diagnosis images of lesion #12 (adenoma), with the photodynamic diagnosis examination times of 0, 60, 180, and 600 s, respectively. Images (b1–b4) show the results of method one overlaid on (a1–a4), respectively, in which tumor regions identified by method one are depicted in red. Images (c1–c4) show results of method two overlaid on (a1–a4), respectively, in which tumor regions determined by method two are depicted in brown, and non-tumor regions are depicted in green.

Figure 10

Effect of photobleaching on the area efficiencies: (a) Points in the photodynamic diagnosis image of lesion #12 are plotted with corresponding values in LAB color space, for tumor points with observation times of 0 s (red circle) and 60 s (blue circle) and non-tumor points with observation times of 0 s (green cross) and 60 s (cyan cross). Arrows depict the direction of the change from points at 0 s to those at 60 s. (b) Area efficiencies of lesion #12 (adenoma), method one (blue circle), and method two (orange triangle), as a function of observation time length.