Spatially Resolved Fibre‐Optic Probe for Cervical Precancer Detection Using Fluorescence Spectroscopy and PCA‐ANN‐Based Classification Algorithm: An In Vitro Study

• Published in: Journal of biophotonics 2024-11, Vol.17 (11), p.e202400284-n/a
• Main Authors: Shukla, Shivam, Deo, Bhaswati Singha, Nemichand, Singh, Pankaj, Pandey, Prabodh Kumar, Pradhan, Asima
• Format: Article
• Language: English
• Subjects: Algorithms; artificial neural network; Artificial neural networks; Cancer; Cervical cancer; Classification; epithelial cancer; Epithelium; Excitation spectra; Female; Fiber lasers; Fiber Optic Technology; Fibers; fibre‐optic probe; Fluorescence; Fluorescence spectroscopy; Humans; Neural networks; Neural Networks, Computer; Optical Fibers; Optics; Precancerous Conditions - pathology; Principal Component Analysis; Principal components analysis; Semiconductor lasers; Signal classification; spatially resolved fluorescence spectroscopy; Spectrometry, Fluorescence; Spectroscopic analysis; Spectroscopy; Spectrum analysis; Uterine Cervical Neoplasms - pathology
• ISSN: 1864-063X; 1864-0648; 1864-0648
• DOI: 10.1002/jbio.202400284

ABSTRACT Cervical cancer can be detected at an early stage through the changes occurring in biochemical and morphological properties of epithelium layer. Fluorescence spectroscopy has the ability to identify these subtle changes non‐invasively and in real time with good accuracy in comparison with conventional techniques. In this paper, we report the usage of a custom designed spatially resolved fibre‐optic probe (SRFOP), which consists of 77 fibres in two concentric rings, for the detection of cervical cancer using fluorescence spectroscopy technique. The aim of this study is to classify different grades of cervical precancer on the basis of their fluorescence spectra followed by a robust classification algorithm. Fluorescence spectra of 28 cervical tissue samples of different categories have been recorded using six detector fibres of FOP at different spatial locations with the source fibre (SF). A 405 nm laser diode source has been utilised to excite the samples and a USB 4000 Ocean Optics spectrometer to collect the output spectra in the wavelength range 400–700 nm. Principal component analysis (PCA) was applied to the collected spectra to reduce the dimensionality of the data while preserving the most significant features for classification. The first 10 principal components, which captured the majority of the variance in the spectra, were selected as input features for the classification model. Classification was then performed using an artificial neural network (ANN) with a specific architecture, including an input layer, hidden layers, and a softmax activation function in the output layer. Experimental and classification results both demonstrate that proximal fibres (PFs) perform better than distal fibres (DFs) in capturing the discriminatory features present in the epithelium layer of cervical tissue samples as PF collect most of the signal from the epithelium layer. The combined approach of spatially resolved fluorescence spectroscopy and PCA‐ANN classification techniques is able to discriminate different grades of cervical precancer and normal with an average sensitivity, specificity and accuracy of 93.33%, 96.67% and 95.57%, respectively. The study reports the usage of a spatially resolved fibre‐optic probe with multiple fibres for fluorescence spectra collection from cervical tissue samples, addressing data imbalance problem using SMOTE, dimensional reduction and feature extraction using PCA, and ANN for classification purpose.