Early Prediction of Oral Precancerous Lesions Using Artificial Intelligence

Pradeep kumar  Rathinavelu; Vyshali  Pillai; Pradeep Kumar  Yadalam; Sri Shivasankari  thilagar; Amir  Raee; Artak  Heboyan

doi:10.18502/fid.v23i14.21568

Pradeep kumar Rathinavelu Department of Public Health Dentistry, Saveetha Dental College, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, India
Vyshali Pillai Department of Public Health Dentistry, Saveetha Dental College, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, India
Pradeep Kumar Yadalam Department of Periodontics, Saveetha Dental College, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, India
Sri Shivasankari thilagar Department of Periodontology and Implantology. Priyadarshini Dental College and Hospital, Thiruvallur Taluk & District, India
Amir Raee Dental Implant Research Center, Dentistry Research Institute, Tehran University of Medical Sciences, Tehran, Iran
Artak Heboyan Department of Research Analytics, Saveetha Dental College and Hospitals, Saveetha Institute of Medical and Technical Sciences, Saveetha University, Chennai, India

DOI: https://doi.org/10.18502/fid.v23i14.21568

Keywords: Wavelet analysis; Leukoplakia; Erythroplasia; Oral Submucous Fibrosis; Candidiasis; Artificial Intelligence

Abstract

Objectives: Oral potentially malignant disorders may present as white, red, or mixed red-white lesions and require accurate early recognition. This study evaluated whether texture-analysis features extracted from clinical digital images could distinguish oral precancerous lesions from other oral mucosal lesions and normal mucosa using gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), and wavelet analysis.

Materials and Methods: Sixty-four clinical digital images were selected according to predefined inclusion and exclusion criteria. The dataset included leukoplakia, erythroplakia, oral submucous fibrosis, candidiasis, lichen planus, leukoderma, frictional keratosis, white spongy nevus, and normal mucosa. Regions of interest were extracted from each image, and texture features were derived using GLCM, GLRLM, and wavelet analysis. A support vector machine (SVM) classifier was then used to categorize images as oral precancerous lesions or non-precancerous/normal mucosa.

Results: GLCM yielded the highest classification accuracy (88%), followed by GLRLM (81%) and wavelet analysis (79%). The corresponding sensitivity values were 77%, 64%, and 60%, and the specificity values were 93%, 90%, and 89%, respectively. The positive predictive values were 83% for GLCM, 75% for GLRLM, and 75% for wavelet analysis.

Conclusion: GLCM-based texture features provided the best diagnostic performance in this dataset. These image-analysis methods may be useful as non-invasive adjuncts to conventional clinical examination and histopathological diagnosis; however, larger datasets and external validation are required before clinical implementation.