| MS TSA Meeting

Name of the Speaker: Mr. Antony Raj (EE21S070)
Guide: Mohanasankar Sivaprakasam
Venue: Online
Online meeting link: https://meet.google.com/xko-pbty-bym
Date/Time: 4th July 2024 (Thursday), at 4:30PM to 5.00PM
Title: Oral Cancer Detection through Endoscope-based Multispectral Imaging

Abstract :

In the field of oral medicine, detecting oral cancer at its onset poses a significant challenge due to its resemblance to healthy tissue, underscoring the need for innovative methodologies. Multispectral imaging, known for its ability to capture images across multiple spectra, stands out by providing in-depth information compared to traditional imaging solutions. While dental professionals have direct access to the oral region, visualizing specific areas can be challenging without suitable tools. This system introduces an innovative method that utilizes endoscope-based multispectral imaging technology for analyzing oral regions affected by cancer. Our multispectral imaging system comprises an imaging sensor with a resolution of 270 x 510 pixels across 16 spectral bands (460 nm to 600 nm), attached to a rigid scope (4 mm - diameter, 0-degree tip angle, 175 mm - length) using a lens coupler. A study was conducted on 10 subjects diagnosed with oral cancer, collecting data with and without endoscope, with a total of 320 images. The acquired images of cancerous tissue was analysed in two ways to identify their spectral signatures. In the first analysis we compared the accuracy of the Spectral Angle Mapper (SAM) algorithm for detecting malignant lesions. The SAM detected output was validated with the ground truth (manually annotated by a dentist) and the results showcased an overall accuracy score of 64.14%, with patient-specific accuracy ranging from as high as 92.84% to as low as 29.53%. In the second analysis, we examined the distance between the spectral signatures of malignant and non-malignant tissue from the images captured using endoscope and non-endoscope methods. The mean difference was 0.7582 for the endoscope method and 0.4218 for the non-endoscope method, indicating the superior detection capability of endoscope based method. Our findings indicate that information from the endoscope-based multispectral imaging system is an effective method for detecting malignancy, especially when combined with SAM and spectral signature analysis, outperforming non-endoscope-based and traditional methods. Overall, our study demonstrates that endoscope-assisted multispectral imaging for oral cancer detection is highly effective and holds significant potential for the design and development of an oral cancer diagnostic device.