2.2.1. Image Processing

For this stage, with the only purpose being to identify carious lesion areas from images, we merged VNC class and C class into the “Caries” category (ICDAS Code 1-6), while NSC class remained with the same number of images, but we named it the “Non-Caries” category (ICDAS Code 0). Then, the image processing algorithm was applied for both categories, “Caries” and “Non-Caries”. The OpenCV software package [22] was applied to perform the image processing procedure. The process is described as follows:

Step 1. ImageEnhance Module and Histogram equalization are applied to the whole dataset for contrast enhancement.

Step 2. Convert all color images to grayscale images (gray_img).

Step 3. Convert grayscale images to binary images (binary_img).

Step 4. Apply Contour Finding algorithm to find the largest boundary of tooth image in binary_img. Use information from accepted contour to create a mask_binary.

Step 5. The relative area outside the contour in the mask_binary is filled with the value 255 to create thick_mask.

Step 6. Tooth contour image (tooth_contour_img) = 255—thick_mask.

Step 7. Then, the image of the carious lesion area (lesion_area_img) is created: lesion_area_img = tooth_contour_img—binary_img.

Step 8. The Contour Finding algorithm is applied again for lesion_area_img to find the largest boundary of lesion and mask_lesion is created from the accepted contour.

Step 9. The relative area outside the contour in the mask_lesion is filled with the value 255.

Step 10. The relative area inside the contour in mask_lesion is replaced with the information of original image.

The performance of the proposed image processing method for caries detection with both “Caries” and “Non-Caries” categories is demonstrated in Figure 3. The post-processing image of a carious lesion area was correctly extracted and shown in Figure 3a with C class and in Figure 3b with VNC class, whereas, by the same process, the output image of “Non-Caries” category did not display any information on the presence of occlusal caries (see Figure 3c).

Proposed image processing method for caries detection. (a) Cavitated class; (b) Visually Non-Cavitated class; (c) No Surface Change class.