Page 139 - AIH-2-3
P. 139

Artificial Intelligence in Health                                   Early detection of CIN/cervicitis using ML



              The most common symptoms reported by patients    could detect subtle cellular changes that might be missed
            enrolled in the study (Table A1) were abdominal pain and   by human experts. ML-powered digital diagnosis and
            abnormal bleeding. Fewer patients had spotting, abnormal   ML-based screening would enable remote diagnosis in
            discharge, and irregular menstrual cycles. The presence of   low-resource settings with limited access to specialists.
            blood or moisture in the cervix could affect the DR image   In this study, an overall diagnostic accuracy of 94%
            ratio values. These cervical conditions may cause light to   was achieved using the DR image ratio R610/R540
            reflect off the surface, resulting in specular reflection. The   algorithm for the discrimination of malignant tissues from
            resulting ratio values from specular reflection can distort   normal tissues. However, the data derived using collagen
            DR measurements. Therefore, when a patient has bleeding   fluorescence could not lead to a confirmative diagnosis.
            or experiences bleeding on touch, the DR ratio values   Alternative markers relying on protoporphyrin IX
            may be misleading. This limitation can be minimized by   fluorescence with excitation at 405 nm could be utilized, as
            cleaning the cervix before imaging. Another limitation of   in OralScan, to improve screening accuracy. Furthermore,
            this study is the low number of malignant cases. Further   the intensity and uniformity of illumination could be
            multicenter studies are planned to improve the algorithm   enhanced by the addition of more LEDs.
            and develop a grade-specific classification.
                                                                 The  mortality rates  due  to cervical  cancer  could be
              During conventional cervical check-ups, most women   minimized by utilizing CerviScan in large-scale screening
            who undergo a Pap smear test do not return to collect the   programs across the country. We believe that early
            reports, which are usually available in 2 weeks for further   detection of cervical cancer would reduce healthcare costs
            follow-up. The non-availability of real-time screening   and minimize the number of patients lost during follow-up,
            results is a significant drawback for screening modalities   as often occurs with traditional screening methods such as
            such as Pap smears or colposcopy-directed biopsies. The   Pap smear and HPV deoxyribonucleic acid tests, which
            ability of the multimodal imaging system (CerviScan) to   require a waiting period of 1  week to get the screening
            provide real-time results, with the help of an ML algorithm,   results. The potential of the device to screen and detect
            presents a strong rationale for adopting this novel   cervical cancers at the point-of-care in real time, along
            technology in screening women above 30  years of age.   with its capability for biopsy guidance – without the need
            CerviScan can help to identify, locate, and treat CIN, while   for visual inspection with acetic acid/visual inspection with
            minimizing its possibility to progress to higher grades of   Lugol’s iodine staining – will improve patient compliance
            CIN and reducing mortality. In addition, the capability   and its effectiveness as a novel modality for the early
            of CerviScan to locate the most malignant site for biopsy   direction of cervical cancers.
            enhances its value by providing a more accurate pathology.
            The real-time image ratio (R610/R545 and R610/R575)   Acknowledgments
            analytics and PCM display of autofluorescence, as well as
            HbO  absorption maps, help in the speedy assessment of   The authors are grateful to acknowledge the support
                2
            malignant status and/or cervicitis.                received from the Ethics Committee at the Kasturba
                                                               Medical College (KMC), Manipal, India. The  authors
            5. Conclusion                                      would like to thank Prof. Dr. Stanley Mathew, Department
                                                               of Surgery, KMC, for his invaluable guidance and advice.
            Quantitative  detection  of  cervical  malignancies  The cooperation and support received from the enrolled
            without  the  need  for  tissue  incision  holds  great  clinical   patients was commendable.
            significance. The algorithm developed for non-invasive
            multispectral widefield imaging using CerviScan provides   Funding
            an opportunity to objectively screen and detect cervical
            cancer in real time, while also assisting the gynecologist   The project was partially supported through a grant received
            in locating the malignant site for biopsy. The cloud-  under the ELEVATE 100 program of the Government of
            based  platform  approach  will  present  a  more  accessible   Karnataka in 2017.
            and cost-effective  solution for  cervical cancer  screening.   Conflict of interest
            The data from different centers, including patients with
            diverse age groups, menstrual phases, and gynecological   The authors declare that they have no competing interests.
            histories, would refine the algorithm over time. As there   Author contributions
            are changes in the cervical anatomy during the menstrual
            span of women, age-specific algorithms would enhance   Conceptualization: Subhash Narayanan
            the diagnostic accuracy. ML algorithms can outdo the   Formal analysis: Phebe George, Rinoy Suvarnadas
            interobserver  variability  among  cytology  results  and   Investigation: Phebe George, Rekha Upadhya


            Volume 2 Issue 3 (2025)                        133                               doi: 10.36922/aih.8527
   134   135   136   137   138   139   140   141   142   143   144