Detection of white spot lesions by segmenting laser speckle images using computer vision methods

Author(s)

Gavinho, Luciano G. ^[1] ; Araujo, Sidnei A. ^[1] ; Bussadori, Sandra K. ^[1] ; Silva, Joao V. P. ^[1] ; Deana, Alessandro M. ^[1]

Affiliation

^[1] Univ Nove Julho, Comp Sci, Campus Vergueiro, Rua Vergueiro 235, BR-04716003 Sao Paulo, SP - Brazil

• Document type: Journal article
• Source: Lasers in Medical Science; v. 33, n. 7, p. 1565-1571, SEP 2018.
• Web of Science Citations: 1

Abstract

This paper aims to develop a method for laser speckle image segmentation of tooth surfaces for diagnosis of early stages caries. The method, applied directly to a raw image obtained by digital photography, is based on the difference between the speckle pattern of a carious lesion tooth surface area and that of a sound area. Each image is divided into blocks which are identified in a working matrix by their chi(2) distance between block histograms of the analyzed image and the reference histograms previously obtained by K-means from healthy (h\_Sound) and lesioned (h\_Decay) areas, separately. If the chi(2) distance between a block histogram and h\_Sound is greater than the distance to h\_Decay, this block is marked as decayed. The experiments showed that the method can provide effective segmentation for initial lesions. We used 64 images to test the algorithm and we achieved 100% accuracy in segmentation. Differences between the speckle pattern of a sound tooth surface region and a carious region, even in the early stage, can be evidenced by the chi(2) distance between histograms. This method proves to be more effective for segmenting the laser speckle image, which enhances the contrast between sound and lesioned tissues. The results were obtained with low computational cost. The method has the potential for early diagnosis in a clinical environment, through the development of low-cost portable equipment. (AU)

• FAPESP's process: 15/25180-7 - Computer vision coupled with coherent illumination applied to diagnostic procedures in odontology
• Grantee: Alessandro Melo de Ana
• Support Opportunities: Regular Research Grants