Page 134 - AIH-2-3

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

images, with reference to the pseudocolor intensity maps for clarity, and if required, the images were recaptured. The
of autofluorescence and the DR images displayed on the imaging of the entire cervix enabled clinicians to identify
screen. The software detected and marked the site with the the optimal areas for further investigation.
highest ratio value in the abnormal area or ROI. A biopsy The cervical smear was collected from the patients
was obtained from this site, and the results were correlated referred for Pap smear testing using a cervical broom and
with the histopathology findings. sent for cytology examination. Women with clinically
Before the initiation of the imaging procedure, the suspected cervical malignancy underwent either a
camera sensor was calibrated to the ambient light conditions conventional or colposcopy-guided cervical biopsy.
using a tissue phantom. The device was positioned at a The site for the biopsy was decided based on the visual
working distance of 15 cm from the tissue phantom, and impression and the gynecologist’s experience. The cervix
the calibration icon in the software was clicked. The four was anatomically labeled using clock positions, with the
sets of LEDs blinked sequentially until the calibration was anterior tip at the 12 o’clock position and the posterior tip
successfully completed. During the calibration process, the at the 6 o’clock position. The cervical arteries and veins
shutter speed of the imaging camera was adjusted relative run parallel in 3 and 9 o’clock positions, and biopsies were
to the LED intensity to accommodate the dark room generally avoided in this area.
conditions to provide the same levels of image contrast, 2.4. Data processing
as verified in the captured image following the calibration
procedure. A calibration window appeared after the Figure 2 illustrates the data processing workflow of
calibration was successfully completed. The device needs CerviScan. In the first step, the camera captured four
to be recalibrated when there is a change in the room light monochrome images. In the pre-processing step, the
intensity. software created pixel-by-pixel ratio images of DR-recorded
images (R610/R545 and R610/R575). The PCMs of these
2.2. Study design and patient enrollment ratio images, along with the PCM of the F375 (fluorescence)
The clinical study was conducted at the Department of image, were displayed from blue to red, with increasing
Obstetrics and Gynecology, Kasturba Medical College, grades of malignancy. For feature extraction, the areas of the
Manipal, India, after obtaining approval from the cervix with clinical significance were identified by adjusting
Institutional Ethics Committee (Ref no. 152/2020). This the color bar in the PCM image of tissue autofluorescence
trial was also registered with the Clinical Trial Registry of and the DR ratio image (R610/R545), followed by marking
India (REF/2020/08/036212). the ROI on either the fluorescence or DR ratio images. In
the post-processing, the software identified the highest ratio
Study participants were enrolled according to the value in the marked ROI for abnormal tissues and an average
study’s inclusion criteria. Women who were referred of the values for normal tissue. The program identified the
for a Pap smear test and those with clinically suspicious most malignant site within the ROI by locating the pixel
lesions in their cervix were included in the study. Women with the highest ratio value.
with transmittable diseases and those who had previously
received treatment for cervical cancer were excluded from 3. Results
the study. Signed informed consent was obtained from A total of 109 patients were recruited for this study. Out
all participants before the initiation of any examination of the study population, 42 patients with histopathology
procedures. results were included in the analysis (Table A1). The results
of histopathology and cytology were correlated with the
2.3. Data acquisition
R610/R545 ratio values of the biopsy site. The patients who
The patient was asked to lie down in the supine lithotomy underwent hysterectomy following a negative Pap smear
position. After cleaning the vaginal area with a soft cotton test were also included as healthy subjects. The results
swab, the gynecologist inserted a speculum for proper of histopathology or cytology were correlated with the
visualization of the cervix. A colored photo of the cervix R610/R545 ratio values of the biopsy sites. The processed
was taken as a reference. The CerviScan device was held DR image ratio R610/R575 was analyzed to discriminate
steadily and moved back and forth to obtain a focused cervicitis from normal tissues.
image on the screen. The capture switch on the device
was pressed, and four monochrome images of the cervix 3.1. Case studies
were recorded and stored in the computer following The severity of the disease in the cervix was assessed based
sequential illumination at 370 nm, 545 nm, 575 nm, and on the percentage increase in the DR ratio R610/R545
610 nm wavelengths. The captured images were reviewed value compared to that of the normal cervix. Low DR

Volume 2 Issue 3 (2025) 128 doi: 10.36922/aih.8527