Consistent with previous studies, patients diagnosed with COD had an average age of 41.28 ± 13.07 years and were predominantly women (88%). This finding is similar to the report that COD primarily affects Asian women, especially those in their 40s and 50s [10]. These findings support previous studies showing that hormonal factors have a complex effect on the development of the disease [11].

The findings of this study revealed a significant difference in texture analysis between COD and PC patients. Among 279 texture features, COD and PC revealed significant differences in 10 GLCM features. The 10 features included entropy, sum of average, sum of entropy, and difference of entropy.

Open-access MaZda version 4.6 offers three different analysis methods: statistical, model-based (fractal or stochastic model), and image transform (Fourier, Gabor, or wavelet transforms) [12]. The GLCM is a second-order statistical method used for texture analysis in 2D images, but it has also been extended to 3D surfaces. GLCM captures different combinations of pixel brightness values in an image and can compute features for single orientations or combine them for direction-independent analysis. However, a significant drawback is the computational cost, which can be addressed by combining GLCM with the Sobel operator to reduce processing time [13].

Specific orientations or directions are considered when computing GLCM features. Texture feature names include four directions and a distance from one to five. For example, we would represent a distance of two in a horizontal direction as (2,0), and a distance of two in a vertical direction as (0,2) [12]. Both entropy (the degree of disorder between pixels in an image) and sum of entropy (the disorganization of the sum distribution of gray shades) increased in CBCT images of patients with COD. This is thought to be due to a mixture of fibrous stroma with loose fibroblasts and collagen with mineralized tissue in COD. Large differences between adjacent pixels increase non-uniformities, and thus the sum of average (a mean of the distribution of the sum of gray shades) increases in CBCT images of patients with COD. In the CBCT images of patients with COD, the difference of entropy, which represents a disorganization of the gray shade difference, increased. This is because the periapical bone of COD patients is replaced by fibrous tissue, including bone and cementum like tissue, resulting in different shades of gray for each tissue in the CBCT image.

Although the sample size was very small, texture analysis showed a statistically significant difference between osteolytic-stage COD and PC, suggesting that the qualitative radiologic assessment was similar but the quantitative assessment was different between the two. The 10 selected variables showed a greater difference between PC and osteolytic-stage COD than the difference between the two COD stages, suggesting that the variables are diagnostically valuable indicators to differentiate COD from PC using texture analysis. This difference is thought to be due to differences in the histologic composition within the lesions. COD has fibrous tissue containing histologically woven bone or cementum-like tissue, whereas PC contains semi-solid necrotic material and cellular components due to liquefaction of epithelial cells [1, 14]. Since the 10 texture features were selected to distinguish between CODs and PCs, we believe that the results would be different if the texture features were selected to distinguish between osteolytic-stage and cementoblastic-stage CODs.

Depending on the stage of development of the lesion, CODs exhibit different radiologic characteristics [15, 16]. Qualitative assessment of CBCT findings by human eyes is possible between COD and PC. However, as noted in many case reports, misdiagnoses often occur, and qualitative assessments that rely solely on the human eye can be unreliable [3, 17, 18]. In fact, owing to the difficulty in differential diagnosis of the two, 7 of the 25 patients diagnosed with COD in this study had to undergo biopsy for confirmatory diagnosis. MDCT is believed to be more accurate than CBCT at very osteolytic stages; however, it is not the first choice because of the high radiation dose [19]. Therefore, texture analysis using CBCT is a non-invasive method that can compensate for these shortcomings, provide a quantitative assessment, and prevent endodontic or surgical intervention.

The ROC analysis of this study yielded 10 cutoff values for differentiating PC and COD, calculated by the Youden index. It also provided a high level of accuracy, ranging from 0.88 to 0.96. This indicates that CBCT texture analysis helps differentiate between COD and PC. Therefore, the analysis of CBCT texture features can provide a quantitative assessment of COD to prevent unnecessary endodontic or surgical intervention, thereby reducing the risk of infection from biopsy.

CBCT evaluation has recently increased and is primarily used for COD diagnosis, as it allows three-dimensional imaging with minimal distortion, high spatial resolution, and no overlap [20]. A comprehensive evaluation of demographic, clinical, radiologic, and follow-up information is typically used to diagnose COD [11, 21]. A biopsy is contraindicated in these lesions because of the deposition of cementum-like tissue and reduced local vascularity, which increases the risk of infection [10, 11, 15, 22]. Generally, COD does not require treatment because routine radiological follow-up is the preferred therapeutic modality [23]. Therefore, many previous studies have used CBCT rather than invasive histologic biopsy to qualitatively diagnose COD [24, 25]. The diagnosis of COD, which was previously based on a qualitative evaluation using CBCT, can be quantitatively assessed using texture analysis as a new non-invasive method to further differentiate it from PC.

Owing to the large number of scattered rays that enter the detector, CBCT is typically not appropriate for calculating quantitative texture features [26]. Even if the CBCT scan time is shorter than the CT scan time, the texture parameters can be affected by motion artifacts. However, some studies using CBCT texture analysis for maxillofacial diseases have been published recently [8, 27, 28]. De Rosa et al. performed CBCT texture analysis to distinguish between a radicular cyst and a periapical granuloma [27]. Goncalves et al. performed CBCT texture analysis to detect furcal lesions [8]. Cost et al. performed CBCT texture analysis to analyze alveolar bone features to confirm implant stability in the maxillary edentulous area [28]. In addition, according to earlier research, some CBCT radiomics features can be used as quantitative indicators for texture analysis [29].

A limitation of this study is that texture analysis was conducted using a small sample size. This is because the lesions are usually asymptomatic and are often discovered by chance on routine dental radiographs performed for other reasons. Future studies should evaluate osteolytic-stage COD using a larger sample size.