Two-Photon Microscopy based Oral lesion classification using AI models
Edwin Pious1, Manikanth Karnati1, Jackson Rodrigues2, Gagan Raju1, Guan-Yu Zhuo2, Nirmal Mazumder1; 1Department of Biophysics, Manipal School of Life Sciences, Manipal Academy of Higher Education, Manipal, India; 2Institute of Biophotonics, College of Biomedical Science and Engineering, National Yang Ming Chiao Tung University, Taipei, Taiwan
Abstract
Early diagnosis of oral cancer is crucial for improving patient survival. However, clinical diagnosis relies upon visual inspection and staining, followed by staging and grading by a pathologist, which is often influenced by subjectivity, leading to delayed detection and treatment. In this study, we employed a custom-built two-photon microscope to acquire label-free images of oral tissue sections. Two-photon fluorescence (TPF) and second harmonic generation (SHG) signals were simultaneously captured to obtain complementary cellular metabolic information from endogenous fluorescent molecules and structural information, particularly from stromal collagen. Deep feature representations systematically extracted from the images using the ConvNeXt architecture. These features were then subsequently classified using a multiclass support vector machine (SVM) employing a radial basis function (RBF) kernel. The proposed model demonstrated robust performance and achieved class-wise F1-scores of 0.78 for oral potentially malignant disease (OPMD), 0.91 for malignant, and 0.99 for control, with an overall accuracy of 90%. Furthermore, this automated framework eliminates the need for ex-vivo staining, significantly speeding up evaluation times. These findings demonstrate the potential of combining label-free two-photon imaging with artificial intelligence for rapid and objective screening of oral lesions.
Speaker
Edwin Pious
Department of Biophysics, Manipal School of Life Sciences, Manipal Academy of Higher Education, Manipal, Karnataka, India
India
Discussion
Ask question
