Materials and methods

Study design and setting

This current research was undertaken as a prospective cross-sectional exploratory study in the Department of Forensic Medicine and Toxicology at King George’s Medical University (KGMU), Lucknow, Uttar Pradesh, India. In this study, the aim was to assess the morphometric as well as gray-level texture features of postmortem injuries using standardized digital image analysis techniques. This study was planned as an exploratory study to generate hypotheses rather than validate diagnoses. Standardized image acquisition and quantitative image analysis were stressed in this research.

Ethical approval

Ethics clearance was provided by the Institutional Ethics Committee of King George’s Medical University, Lucknow (IEC No.: 104th ECM II B-PhD/P5). Anonymization was done for all medico-legal images used in this study. No personal details were stored in the data bank. The images were used for educational and research purposes according to medico-legal norms set by the institution.

Study population

The study sample consisted of medico-legal autopsy cases reported to the Department of Forensic Medicine and Toxicology at King George’s Medical University, Lucknow, during the study period. The unit of observation in this study is the image of injury, not the autopsy case. A total of 383 autopsy images depicting injuries that met the set criteria were used for quantitative image analysis. Images selected for analysis depicted various forms of injury as reported in medico-legal autopsies. In cases with multiple injuries in a single autopsy, if each injury met the inclusion criteria, more than one image per case could be included. This was because each injury had to be observed independently. The present study aimed at exploring image analysis; hence, all images meeting the inclusion criteria during the study period were used.

Sampling technique and sample size

Consecutive sampling methodology was used in this study. Any image showing injuries meeting the eligibility criteria as well as having image quality criteria was included. Because the focus of the study was exploratory in terms of assessing morphometric and grayscale properties of injuries sustained after death, the injury image was taken as the basic unit of analysis. A total of 383 eligible images of injuries were used for analysis. If several images were obtained from the same autopsy case, those images that depicted different injuries were included. As a result of the exploratory nature of the research and lack of data for forensic morphometric studies using imaging, no sample size calculations were conducted.

Inclusion criteria

We included medico-legal autopsies where there were observable injuries to the body with easily identifiable margins to allow for proper measurement. Only injuries whose margins could be clearly determined were used. For inclusion in the study, good pictures of the body and an ABFO No. 2 forensic scale had to be present in the picture.

Exclusion criteria

The study left out cases where injuries were too decomposed or badly burned to be assessed properly. We also did not use photos that were fuzzy, poorly lit, or just plain distorted. Any pictures where the injuries were changed due to stuff that happened after death were not used either. In addition, we excluded anything without a scale for proper measurement. Lastly, if environmental damage messed with how the images could be interpreted, those were removed as well.

Operational definitions

Abrasion was described as a superficial injury where there is a loss of integrity or damage to the epidermis but no substantial amount of subcutaneous bleeding. Contusion was described as a closed soft-tissue injury where there was subcutaneous bleeding but no disruption to skin integrity. Because the subgroups were not adequately sized, gunshot and electrocution injuries were grouped under “other injuries."

Image acquisition protocol

To control the effect of confounding variables, the forensic photography process was strictly standardized for this study. For the study, all the photographs used had been taken using the same Nikon DSLR camera (Nikon Corporation; Tokyo, Japan) with consistent imaging conditions. In particular, the Nikon camera was always placed at a 90° angle perpendicular to the injury surface, at the same working distance where possible. Consistent camera settings were adopted (ISO 100-200, aperture f/8-f/11, white balance set to fixed, and exposure). Each photograph had the ABFO No. 2 forensic scale placed in it. The exclusion criteria included blurred photographs, out-of-focus photographs, photographs with shadows, reflections, overexposure, underexposure, misplacement of scale, and an oblique angle of the camera. To reduce variability resulting from image processing, all images went through the same process of calibration, grayscale conversion, and normalization. Besides, highly decomposed bodies, burned wounds, and wounds with vague margins were excluded from the study.

Image processing and analysis

Image processing in digital images was conducted using the ImageJ software (version 1.53, National Institutes of Health, Bethesda, MD, USA). All obtained images were subjected to uniform preprocessing before being quantitatively analyzed. Preprocessing consisted of conversion into grayscale, calibration by ABFO No. 2 forensic scale, and normalization of intensity values. Standardized procedures for image preprocessing were used for all images in the study. The standardized forensic photography and ImageJ-based image analysis workflow is illustrated in Figure 1.

Figure 1: Standardized forensic photography setup showing injury documentation using a DSLR camera with ABFO No. 2 scale for calibration under controlled conditions.Standardized forensic photography and ImageJ-based image analysis workflow. (A) Documentation of postmortem injury using a DSLR camera under controlled photographic conditions. (B) Injury photograph showing placement of the ABFO No. 2 forensic scale for calibration. (C) Manual delineation of the region of interest (ROI) in ImageJ software for morphometric analysis. (D) Calibration and quantitative measurement of injury dimensions using ImageJ software following ROI selection.

Region of interest (ROI) selection

The selection of region of interest (ROI) was done manually based on standardized criteria set forth beforehand. The edges of the injury were precisely demarcated such that uninvolved tissue adjacent to the injury was not incorporated. In order to minimize subjectivity on the part of the observer, the selection process was done independently by two observers who had knowledge of ImageJ image analysis methods. Differences in opinion among observers were reconciled before measurements were recorded.

Observer reliability assessment

Before conducting any image measurements, both observers were equally trained on how to choose ROI, define the borders of injuries, calibrate the images, and measure using the ImageJ software. A standard operating procedure (SOP) was strictly followed during image measurements. Inter-observer reproducibility was evaluated using the intraclass correlation coefficient (ICC), which proved highly reliable for area (ICC=0.88), perimeter (ICC=0.84), and solidity (ICC=0.81).

Parameters evaluated

The study investigated both morphometric and gray-level textural measurements derived from digital image processing. The morphometric measurements consisted of length, width, area, perimeter, and solidity, all chosen for evaluating the geometry, size, and borders of injuries. On the other hand, gray-level textural measurements included minimum gray level, maximum gray level, median gray level, and SD of gray levels. These measurements were considered for assessing pixel intensity and tissue heterogeneity within the selected injury areas.

Confounding variables

A number of steps were taken to minimize the influence of possible confounding variables on the results of the study. Similar photo-taking conditions, same camera settings, similar calibration processes, and same image processing methods were applied to all cases. All cases that showed advanced decomposition, burning, environmental deterioration, non-defined injury borders, and poor images were rejected. Only those injuries that had defined borders and were adequate for morphometric analysis were used. While it was impossible to rule out all the possible effects of anatomical region, skin color, time since death, and natural heterogeneity of an injury, some steps were taken toward minimizing their influence.

Data management

All the metrics that were extracted through morphometrics and texture analysis were documented using Microsoft Excel (Redmond, Washington, USA) sheets, where the cases were assigned anonymous identification numbers to ensure confidentiality. This database was validated for any missing data, data entry errors, or inconsistencies before undertaking any statistical analysis.

Statistical analysis

Statistical analysis was conducted using IBM SPSS Statistics v. 26.0 (released 2018; IBM Corp., Armonk, NY, USA) and R software v. 4.3 (R Foundation for Statistical Computing, Vienna, Austria). Continuous data are summarized using mean ± SD, median, interquartile range (IQR), and minimum-maximum range where appropriate. Statistical differences between morphometric and grayscale texture parameters across the injury groups were analyzed using one-way analysis of variance (ANOVA) with Bonferroni post-hoc adjustment. Correlations were determined using Pearson's correlation coefficient analysis. The level of multicollinearity was assessed by the variance inflation factor (VIF). Binary logistic regression was used to find the best predictors of injuries, while the performance of the regression model was checked using the Omnibus test, Nagelkerke R², the Hosmer-Lemeshow test, and the classification accuracy of the model. Receiver operating characteristic (ROC) curve analysis was used to assess the discrimination potential of significant predictors and the logistic regression model; the area under the curve (AUC) was calculated, and sensitivity, specificity, and 95% confidence interval (CI) were determined. Inter-observer reliability was assessed by the ICC. A p-value≤0.05 was set to indicate statistical significance.