The present study shows that hyperspectral imaging allows the CO-Hb concentration in human blood to be determined and qualitatively distinguished from a negative control sample by the contactless analysis of its spectrophotometric characteristics. The validity of the model created is limited by the idealized setup of the experiment in vitro and the limited amount of data for training the model.

Under standardized environmental conditions, the model achieved a prediction error of approx. 4.5 percentage points and maximum errors of less than 11 percentage points. Although this error varies overall within the six-week test, it does not increase linearly with the time elapsed, which is why we assume the method to remain constant over a longer period. If applicable in forensic medical practice, a distinction between pathologically elevated CO-Hb concentration (above 9—15% even in heavy smokers) [7] and physiological values could therefore be possible. This distinction would be even more precise for lethal CO intoxications, in which a concentration of between 30 and 60% CO-Hb is usually measured [13]. Furthermore, our method could fasten the time needed for evaluating COHb concentrations, since currently laboratories can take several days for their results, especially if they are not located near the body´s finding site or the coroner’s office as its often the case in rural areas.

For a practical application in post-mortem examinations (hyperspectral analysis of death spots, mucous membranes or nail beds), the scattering and absorption processes caused by melanin, collagen and water, for example, must also be taken into account, since they are some of the most important chromophores for visible wavelengths, of which the absorption coefficient decreases monotonically with the increasing wavelength [15, 19, 22, 23]. In a study conducted by Bohnert M. et al. [2], in which livor mortis were recorded with a spectrophotometer and analyzed with a stochastic “Monte Carlo” model, the experimental setup featured an immobile, complex recording device and an equally controlled environment. The calculation time of the model was more than 7 days and thus not superior to lab analyses timewise.

For our system, the training of the Lasso regression model only took a few seconds. With the help of the Specim IQ Studio® software, it would be possible to create evaluation procedures (algorithms, so-called "applications") for data processing and analysis and store them in the camera, so that the evaluation of the data takes place in the camera and the result is displayed to the examiner on site. In real blood samples from patients, the hemoglobin concentration and the potential presence of other hemoglobin derivatives—such as deoxyhemoglobin or methemoglobin—and their respective optical properties must also be considered for training of the model. For example, Principal Component Analysis (PCA), Neighborhood component feature selection (NCFS), and Support Vector Machines (SVM) have proofed to be capable of assigning typical spectral bands of individual hemoglobin derivates to spectral classes [3], which could make on-site analysis of the images possible within a few seconds and the result could then be visualized on the camera display.

A key limitation for those would involve situations where illumination conditions are either unstable or of low intensity. In such cases, the mobile light sources employed in this study could provide a uniform distribution of light at the scene. However, to address rapidly changing lighting conditions or extremely confined spaces with minimal movement range, a simple device could be developed. This device would function similarly to the commonly used “matte box” and incorporate a circular light source around the camera lens to ensure consistent illumination and offer some degree of shielding from ambient light interference.

The method presented could also give a second chance in complex cases of forensic investigations, where the integrity of samples yet to analyze has already been decreased by rot or heat-related circumstances. For us, this recently was the case with exhumed material (Postmortem Interval: 15 months) for an appraisal, in which spectrophotometric examination of the cadaveric blood was no longer possible. For the first time, the method developed here was applied to greatly damaged probes and was able to acquire spectrophotometric data, where standard measurement methods delivered only error messages, resulting in the model to at least classify the COHb concentration as “highly elevated “ and therefore provide an additional piece of information for the toxicological evaluation.

The described challenging forensic scenario demonstrates the method’s potential utility but also highlights that its applicability is limited once the level of destruction—such as coagulation, heat exposure, or the presence of burned organic matter—exceeds a certain threshold. The same would apply to skin samples, where burn wounds or charring would significantly compromise the method’s effectiveness.