A clinical study was conducted at the Department of General Surgery, Kasturba Medical College and Hospital, Manipal, between October 2022 and September 2022. The study commenced after obtaining approval from the Institutional Review Board (625/2020) and was registered with the Clinical Trial Registry of India (CTRI/2021/01/030415).
All patients suspected of having gastrointestinal luminal pathology undergoing surgical resection for the diseased segment of the gastrointestinal tract and willing to consent to the study were included. However, patients under 18 years of age and those with non-luminal gastrointestinal pathology were excluded from the study.
Methodology
The consenting participants underwent a standard of care resectional surgery based on the surgeon’s pre-operative evaluation and diagnosis. The resected specimen was handed over to the investigator in the operating room as soon as the surgical procedure was completed. Each specimen was cleaned with water to remove blood and luminal debris. The samples were then oriented with proper labeling and digital photographs were obtained for future reference. The multispectral imaging device was calibrated before recording every patient specimen to the light conditions of the room using the tissue phantom provided. Four monochrome images were captured sequentially by the device under illumination with the four LED light sources, from the most representative predetermined mucosal and serosal areas of the specimen. Each image was captured at a fixed distance from the tumor. Selected lymph nodes were also harvested and imaged as described above. The extracted lymph nodes were histologically examined after individual labeling. Doughnuts and re-excised specimens in procedures using staplers were also photographed. For future reference and identification, corresponding digital colour photographs of the tissues imaged for fluorescence were also gathered and recorded. After the spectral recordings were completed, the samples were immersed in formalin and sent to the laboratory for histological examination.
The proprietary software marks the region of interest based on the pseudo color fluorescence image and DR image R610/R545 ratio values. The software then identified the highest ratio value in the region marked by the researcher.
A pathologist blinded to spectroscopic data evaluated the samples to confirm the presence of malignant and normal tissues using traditional haematoxylin/eosin (H&E) staining. The ratio of R610/R545 was correlated with the histopathology results and presented as a scatter plot diagram.
Specifications of the device
The multispectral imaging device was developed by Sascan Meditech for the bimodal imaging of epithelial tissues. LEDs were positioned in circles around a monochrome high-resolution (5 MPx) camera connected to a zoom lens. A set of ultraviolet-A (375 nm) emitting LEDs was used to examine collagen fluorescence, whereas a set of green (545 nm), yellow (575 nm), and red (610 nm) wavelength LEDs were used to examine the scattering properties of tissue and changes in oxy and deoxy absorption. A customized filter was used to transmit tissue autofluorescence and elastically back-scattered light at 542, 577, and 610 nm into the sensor while preventing excitation light at 375 nm from entering the detection system. Crossed polarizers were used in the optical path to avoid specular reflections from the tissue surface from entering the sensor. The camera was connected to the USB port of a tablet or laptop running 64-bit Windows 10, and image capture and analysis were performed using proprietary software installed on the laptop (Fig. 1A) [6].
The device is focused on the tissue, and multispectral images are captured sequentially using the capture button on software or hardware. The region of interest (RoI) is marked on all images based on the abnormality seen in the processed pseudo-color map (PCM) of the tissue fluorescence and DR ratio images. The software checked the ratio values (R610/R545) and marked the site with the highest DR ratio value in the RoI.
Statistical analysis
A scatter plot diagram of the DR image intensity ratio R610/545 values for different sites was constructed. Cut-offs were drawn at the average ratio values to distinguish normal sites from malignant lesions, normal sites from benign lesions, and malignant lesions from benign lesions (Figs. 2,3,4).
The Z-test ensured that the two groups' mean values were identical. The null hypothesis was that both means were equal. The alternative hypothesis was that the mean number of abnormal sites was more significant than that of normal sites. The null hypothesis was to be rejected if, at a 99% confidence level, the z statistic was less than the z critical value. We have described the qualitative data in terms of the numbers and percentages. Statistical significance was set at P < 0.05.
In classifying normal from malignant, the mean pixel value of the R610/545 image ratio for normal was 1.945, and for malignant, it was 5.032. In the scatter plot of the 199 normal sites, 2 were misclassified as malignant. The table shows the sensitivity, specificity, PPV, and NPV for each pair.
Software
Proprietary software was developed for this study. The software triggers the LEDs in sequential order, captures images, processes the recorded images in real-time, and provides ratio values to determine the tissue status.
The software's home screen has a window with a list of the patients recorded and an option for registering new patients where we can enter patient details such as Unique ID (Fig. 1B). The camera can be operated by the screen menu, followed by calibration of the camera and LEDs with the working dim light conditions. Once the device was calibrated to ambient light conditions, the probe was placed in position and focused for image clarity.
The capture button on the software or the hardware switch on the probe was used to trigger the LEDs and capture the images. At least two separate sites are imaged for each abnormal lesion (depicted as green on the serosal surface and yellow on the mucosal lesion in Figs. 5 and 7), and the site farthest away from the abnormal lesion (depicted as blue on the mucosal surface in Figs. 5 and 7) is chosen to represent the normal area in that resected specimen.
The captured images were processed in the software, and the region of interest (ROI) was marked based on the fluorescence emission pseudo-color image F375 (Figs. 6 and 8). After marking the ROI, the software prompts the selection of whether the marked region is normal or abnormal. The built-in analysis will provide a ratio value corresponding to the changes in oxygenated hemoglobin absorption. Each site's R610/545 ratio value was analyzed to distinguish normal from abnormal tissues.