Our study demonstrated better inter-rater agreement for HFLUT over CUT in both B-mode and M-mode imaging for pleural line and subpleural field characterization in dogs without evidence of pulmonary parenchymal disease and dyspneic dogs. Specifically, M-mode and B-mode clips captured with HFLUT exhibited strong to near perfect agreement for pleural line abnormalities interpretation using our classification system. This suggests that a HFLUT produces images that can be more consistently described and interpreted, which may result in improved diagnostic accuracy compared to a CUT. These findings align with existing human literature indicating that higher frequency linear transducers provide enhanced visualization of superficial structures and pleura, coupled with superior axial and lateral resolution [22].

Moreover, while B-mode imaging primarily assesses anatomical structures and tissue characteristics, M-mode imaging serves as a supportive modality, providing real-time information regarding motion and dynamics [1, 15, 27, 31, 32]. In human medicine, M-mode has been previously validated for the detection of various thoracic abnormalities, including pneumothorax, pleural effusion, and hydropneumothorax, and has also been described for the evaluation of the diaphragm for paralysis and ventilator weaning [22, 30, 31]. Comparatively, in veterinary medicine, M-mode has been largely restricted to the diagnosis of pneumothorax, although clinical studies to support its use are lacking [6, 25]. In our study, we found that there was greater inter-observer agreement amongst reviewers when reviewing the pleural line in M-mode ultrasonographic profile using the HFLUT for dogs with SLIS compared to B-mode using either the HFLUT or the CUT. The one-dimensional representation of motion over time in M-mode likely allows for the identification of specific fields, such as changes in pleural line movement and the presence of vertical artifacts, which are indicative of AIS. These fields were likely easier to recognize and less subjective compared to the real-time, two-dimensional images of B-mode ultrasonography. Consequently, this reduces variability between observers and enhances the consistency and reliability of the assessments, making M-mode a valuable tool in the evaluation of SLIS in veterinary lung POCUS. Further studies enrolling a larger population would be valuable to further evaluate differences between M-mode and B-mode ultrasound to characterize the pleural line.

In current literature, the term ‘vertical artifact’ encompasses a range of phenomena observable on lung ultrasound, the specific criteria for defining and categorizing these artifacts, including B-lines, vary across the literature [9, 11, 21, 22]. In our study, we opted to use the broader term ‘vertical artifact’, in lieu of “B-line”, to reflect the potential for variability in appearance caused by differences in pathology (e.g., fibrosis versus fluid) and technical factors (e.g., machine settings). Thus, this terminology reduces the risk of misclassification and prioritizes inter-rater variability rather than the precise categorization of the artifact’s origin.

For subpleural field characterization, HFLUT’s M-mode also appears to showcase the subpleural field more clearly compared to CUT’s M-mode and B-mode. Furthermore, the strong inter-rater agreement with HFLUT suggests that it might offer a more reliable method of assessing pleural line integrity and the nature of the subpleural fields, which may subsequently improve diagnostic accuracy, although further research is required to support this hypothesis. In light of this, further research to evaluate the pleural line, subpleural fields, and their combinations in dogs presenting in respiratory distress using a HFLUT and a combination of B- and M-mode ultrasonography is warranted.

In our study, we adopted a standardized language derived from human medicine to describe pleural line abnormalities. Recognizing the absence of a consensus for such descriptions in veterinary medicine, we either adapted or formulated definitions from Fischer et al. [15] to establish fundamental characteristics of pleural line abnormalities. While human medicine offers further characterization of pleural lines, applying these inconsistently to veterinary patients poses challenges due to anatomical and breed-associated nuances. Therefore, our approach aimed to tailor definitions to better suit the veterinary context in the hope of creating a more uniform research approach, and diagnostic characterization of AIS.

Using M-mode with a HFLUT in our study, we consistently observed associations between a homogeneous pleural line in either control dogs or those with cardiogenic pulmonary edema (CPE), while a non-homogeneous pleural line was associated with non-cardiogenic alveolar interstitial syndrome (NCAIS). Similarly, the vertical subpleural field identified with HFLUT was associated with either CPE or NCAIS in 16 out of 18 (89%) of the reviewed clips with only two clips of 18 clips (11%), having a discrepancy amongst the reviewers. These findings are similar to those of Singh et al. [32], where in people, the combination of a continuous pleural line and a vertical subpleural field was associated with CPE, and a non-continuous, fragmented pleural line and vertical subpleural field was associated with NCAIS. Furthermore, while the horizontal subpleural field was most frequently associated with control dogs in this patient population, it was also found in several clips within the CPE and NCAIS groups as well. However, it should be noted that horizontal subpleural field identified in patients with CPE and NCAIS were observed in ZOI's without vertical artifacts, which likely explains the discrepancies. Further research is needed to confirm the association identified in this pilot study.

Several limitations should be considered when interpreting the findings of this study. There were a limited number of clips from only nine dogs. While a sample size calculation was performed for statistical purposes, future research with larger sample sizes is warranted to validate our findings and elucidate factors influencing interobserver agreement in ultrasound imaging interpretation (e.g., breeds with various conformations, sizes, sonographer and clip reviewers’ level of experience etc.). For comparison, the study by Fatima et al. (2022) demonstrated higher inter-rater reliability with six human operators analyzing 1035 lung ultrasound videos in COVID-19 patients [14]. However, in veterinary medicine, the use of two expert reviewers allows for consistent, high-quality interpretations essential in an emerging field like veterinary LUS. This approach emphasizes expertise over quantity, reducing variability that could arise with less-experienced reviewers. Additionally, the selection of two reviewers was supported by a power analysis based on other veterinary studies [10, 24, 25, 36, 37], which indicated this would yield statistically meaningful results. Adding more reviewers would require substantial training and resources, potentially overcomplicating the study without improving the reliability of findings. However, the authors acknowledge that future studies could benefit from incorporating a broader pool of evaluators with different levels of experience to further assess consistency across ultrasound assessments, particularly in veterinary settings. This study could also be impacted by variability in imaging technique, although only one investigator (KG) performed all ultrasonographic examinations of the dogs enrolled in the study. Furthermore, the clips attained included videos, which diminished the impact of transducer handling or patient movement (e.g., oblique angling of the transducer during examination) by capturing the lung movement or sliding at various angles throughout the examination. Next, bias on the part of the reviewers to favor one transducer over another could not be blinded as the appearance of the image easily differentiates HFLUT from CUT B-mode clips. Also, the focal point was not specifically adjusted to the pleural line, which may have influenced the findings. However, given that all reviewers assessed the same clips with the same focal position, this limitation is unlikely to have impacted inter-rater variability.

Furthermore, to provide a more robust analysis, a linear mixed model (LMM) would be more appropriate for the statistical analysis of the associations between diseases and clip classifications. LMM can adjust for both fixed effects (e.g., disease category) and random effects (e.g., observer variability). Moreover, our methodology did not involve quantitative counting of vertical artifacts but rather focused on qualitative characterization of pleural and subpleural fields to provide a robust assessment of lung pathology. However, implementing an LMM requires a larger sample size, which was not available in the current study. Future research should aim to collect a larger sample size to enable the use of LMM or similar approaches for a more accurate and reliable analysis of the data, accounting for observer variability.

Finally, an advancing frontier in LUS is the integration of algorithmic and computer-aided diagnostic (CAD) systems to overcome the limitations inherent in manual interpretation. Modern developments suggest that a standardized CAD approach could greatly enhance the identification and quantification of pleural line and subpleural features, increasing consistency across clinicians. Corradi et al. [7] demonstrated that algorithmic LUS, utilizing gray-level co-occurrence matrix (GLCM) and second-order texture analysis, effectively reduces observer variability by emphasizing characteristic features of various respiratory pathologies. This technology could facilitate more uniform LUS assessments, achieving consistent diagnostic scores and precise monitoring metrics, independent of an operator’s expertise. Additionally, CAD tools can streamline LUS assessments by automating pleural line and subpleural feature quantification, thus minimizing observer variability and supporting robust, reproducible scoring frameworks in both research and clinical applications [14].