IADR Abstract Archives
Artificial Intelligence in Early Diagnosis of Precancerous Lesions
Objectives: The diagnosis of Oral pre-cancer remains a challenge particularly in the detection, evaluation and management of early phase alterations. Artificial neural network can be a simple, effective, inexpensive and non-invasive diagnostic and prognostic tool for various premalignant lesions and conditions. The aim of this study is to develop a tool that automates the diagnostic procedure, in diagnosing precancerous lesions and differentiating it from other oral lesions, normal oral mucosa and comparing the results with the actual diagnosis.
Methods: Materials and methods:
In this study the problem is approached by running texture (GLCM- Gray Level Co-occurrence Matrix (GLCM) and GLRLM - Grey Level Run Length Matrix) and wavelet analysis to extract the features in the image. We calibrated values by feeding the precancerous lesions, other oral mucosal lesions and normal mucosa images without any lesions. After the extraction and selection, the features were input into classifier to categorize the images into the oral precancerous images and other oral lesion images. Support Vector Machine (SVM) method is used to categorize lesions.
Results: Results:
A total of 64 precancerous lesions, other mucosal lesions and normal mucosa images measuring 40x40 pixels were transformed into wavelet analysis, GLCM and GLRLM. We obtained up to 79% accuracy using wavelet analysis, 88% using GLCM, 81% using GLRLM. Using the sensitivity and specificity values ROC curve is plotted to evaluate the performance of GLCM, GLRLM and wavelet analysis. Sensitivity and specificity values obtained using wavelet analysis, GLCM and GLRLM.
Conclusions: Conclusion:
In this study, we observed that the texture features can be used to predict precancerous lesions. The strength of this method is that it is simple, non-invasive and quick.
Methods: Materials and methods:
In this study the problem is approached by running texture (GLCM- Gray Level Co-occurrence Matrix (GLCM) and GLRLM - Grey Level Run Length Matrix) and wavelet analysis to extract the features in the image. We calibrated values by feeding the precancerous lesions, other oral mucosal lesions and normal mucosa images without any lesions. After the extraction and selection, the features were input into classifier to categorize the images into the oral precancerous images and other oral lesion images. Support Vector Machine (SVM) method is used to categorize lesions.
Results: Results:
A total of 64 precancerous lesions, other mucosal lesions and normal mucosa images measuring 40x40 pixels were transformed into wavelet analysis, GLCM and GLRLM. We obtained up to 79% accuracy using wavelet analysis, 88% using GLCM, 81% using GLRLM. Using the sensitivity and specificity values ROC curve is plotted to evaluate the performance of GLCM, GLRLM and wavelet analysis. Sensitivity and specificity values obtained using wavelet analysis, GLCM and GLRLM.
Conclusions: Conclusion:
In this study, we observed that the texture features can be used to predict precancerous lesions. The strength of this method is that it is simple, non-invasive and quick.