A Streamlined Telepresent Video Platform for Aerospace Medicine
INTRODUCTION: Improved video transmission is needed for telemedicine in austere or remote ground, maritime, and aerospace environments. A prototype compression algorithm named “V-CRAMMIT” (patent pending) streamlines medical images, improving bandwidth efficiency. This study evaluated the technology by assessing diagnostic designations made by medical clinicians using uncompressed vs. compressed video.
METHODS: An inventory of deidentified videos was selected from a library of recorded MP4 pulmonary ultrasound scans. Videos displayed four lung pathology conditions: Pneumothorax, Pneumonia, Pleural Effusion, and No Finding/No Pathology. Four videos were selected per condition, yielding 16 recordings compressed using V-CRAMMIT. Average file size reduction was 83%. A total of 20 ultrasound clinicians evaluated each video in both uncompressed and compressed format, presented in a randomized 32-trial sequence. In each trial, subjects selected one of the four pathology designations or an “Inconclusive/Unsure” designation. Trials were presented using a mouse interface and virtual dashboard displaying large-format video, the anatomic location of each scan, a replay button, and response buttons for the above designations. Accuracy, response time, and replay frequency were analyzed using repeated-measures ANOVA incorporating Video format (Uncompressed vs. Compressed) and Pathology (four levels) as independent factors.
RESULTS: Performance did not differ between video conditions: Uncompressed vs. Compressed trials showed no significant differences in accuracy, response times, or replays. Pathology imposed significant effects for all measures, with subjects making the most accurate (91% correct) determinations in Pleural Effusion cases.
DISCUSSION: Streamlined video afforded performance equivalent to standard video and could thus enable improved bandwidth efficiency for remote telemedicine settings.
Beer JMA, Ng PC, Wlodarski DC, Tompkins J, Mock J, Mojica A, Clemons M. A streamlined telepresent video platform for aerospace medicine. Aerosp Med Hum Perform. 2025; 96(11):1008–1014.
Telemedicine offers a means to improve medical outcomes and efficiency for providers who have limited resources or are situated in austere environments.1 In a telepresent medical consultation, audiovisual medical information and guidance are exchanged between a local provider offering patient care and a medical specialist at a physically separate, possibly distant facility possessing greater resources.2 Using telemedicine can offer new opportunities to deliver guided therapeutic care in remote settings3,4 and support swifter interpretation of scans in emergency stroke care.5 Telemedicine may improve logistical efficiency, with one study reporting that the need for medical evacuation in U.S. service personnel deployed in Iraq and Syria decreased 57% following implementation of telemedicine services.6 This decrease yielded an estimated reduction of 328 evacuations/10,000 personnel/year, with an attendant savings of 1.27M USD/10,000 personnel/year in 2020.
Information exchange in telemedicine may be asynchronous, whereby audiovisual information is recorded and transmitted for evaluation later. Alternatively, when the need for consultation is urgent or specialized, information exchange may be synchronous, whereby communication between point of care and specialist occurs in real time, culminating in the most advanced form of telemedicine, which is a telepresent surgical interface.7 In both asynchronous and synchronous information exchange, telemedicine depends on bandwidth and information throughput, with network limitations presenting logistical and security constraints. In the Hotker4 study, failures of operability were attributed to reception conditions and transmission rate 69% and 59% of the time, respectively, and while commercial communication networks may seem to provide an adequate audiovisual platform, they often lose signal during teleconferencing without an obvious cause.8 In the asynchronous realm, reduced bandwidth can limit the benefits of telemedicine by delaying consultant responses from a domain of seconds to one of hours,6 while synchronous telemedicine can be hampered critically via latency, video jitter, and interruptions.7 These findings indicate that video transmission must improve if it is to support triage and consultation in bandwidth-constrained conditions, particularly when synchronous communication is required. This need is expected to grow because future telemedicine requirements may emanate from environments that are progressively more hostile to continuous communication, including combat sites, remote field care locations, and ultimately space. Especially in this last arena, where signal delay and weakness are inherent, the potential consequences of interruption are calamitous and the need for bandwidth efficiency acute.
One way to address bandwidth problems is to reduce the quantity of information transmitted, either in each asynchronous exchange of “record/forward” files or throughout the continuous transmission inherent in synchronous telemedicine. This exploratory study was performed to evaluate the clinical effectiveness of a prototype compression algorithm called “Video-Compression using Robust, Adaptive, Multiple Methods and Innovative Techniques” (V-CRAMMIT) which streamlines video files used for medical imaging, reducing their size to speed transmission. V-CRAMMIT, which is the intellectual property of KBR, Inc., and is pending patent approval, offers the potential capability to transmit streamlined images with a fidelity equivalent to that of uncompressed video, yielding advanced bandwidth efficiency. While these gains had been indicated using engineering metrics, they have not yet been verified with human users, especially clinicians possessing expertise reading medical scans.
Here, V-CRAMMIT was deployed in a paradigm designed to emulate asynchronous delivery of medical video content that could be used for triage and diagnosis. A collection of video recordings of point-of-care ultrasound (POCUS) scans showing four different pulmonary pathology conditions was constructed. Recordings were then compressed using V-CRAMMIT. Clinicians with POCUS expertise reviewed recordings in both standard (“Uncompressed”) and V-CRAMMIT (“Compressed”) formats and were asked to identify the pathologies shown. The accuracy and speed of their assessments were recorded and compared across video formats and pathology conditions to determine whether V-CRAMMIT afforded equivalent performance. The study tested the hypothesis that the efficiency gains claimed using engineering metrics would generalize to clinical effectiveness: if clinicians identified pathologies as accurately and swiftly when they viewed medical scans in Compressed vs. Uncompressed format, this would indicate enhanced efficiency whereby V-CRAMMIT could support remote triage and diagnosis with a reduced demand for bandwidth. In military medical applications, this could yield a concomitant decrease in vulnerability to detection and attack.
METHODS
The study protocol was approved in advance by the USAF 59th Medical Wing Institutional Review Board (#FWH20230093E). It was conducted under a waiver from Human Subjects regulation, including required informed consent, because it was determined to impose minimal risk: interventions were not invasive or stressful and datasets and scans were deidentified. The study comprised one test session lasting 30–45 min and was conducted across three sites in San Antonio, TX, including the 59th Medical Wing, University of Texas Health Science Center Emergency Ultrasound Division, and Brooke Army Medical Center.
Subjects
A total of 20 volunteers (7 women, 13 men), ages 29–63 (mean 37.2 yr, SD 7.7 yr), enrolled following e-mail or verbal recruitment. Pilot data were not available, so N was projected from analysis seeking 89% power to detect a 0.75-SD (moderate) difference between Video conditions (alpha = 0.05, two-tailed). Subjects self-reported as licensed Doctor of Medicine (14), Doctor of Osteopathic Medicine (4), or Physician Assistant (2). All reported proficiency in medical image reading and completion of training or experience in pulmonary POCUS as part of their medical practice. Subjects provided demographic information, including specialty and experience, but names were not recorded. Subjects were not required to show credentials because the study was approved as an exempt procedure. Although the protocol required no informed consent documentation, subjects received a description of the procedure, including emphasis that they could withdraw at any time. Subjects were screened for near visual acuity of 20/30 or better. Of the subjects, 6 wore no vision correction, with 10 wearing spectacles and 4 wearing contact lenses.
Enrollment included an eight-trial practice block to familiarize subjects with the procedure. These trials included two examples of each of the four Pathology conditions described below. All practice videos were Uncompressed, so subjects would train on the format offering the most potential information density. This established “Uncompressed” as the default format against which the Compressed format would be compared. Practice trials depicted different patients from the videos used in the experiment, but otherwise resembled the experimental trials described in the Equipment and Procedure sections.
Equipment
V-CRAMMIT comprises an amalgam of video compression and precompression techniques to streamline images and videos while ensuring acceptable visual fidelity and adherence to standards. It optimizes compression by combining lossy and lossless precompression methods. It can be adapted to multiple video or image formats, to various network states (including loading and available bandwidth), and to users’ specified display needs, including recipient display characteristics and area-of-interest circling. V-CRAMMIT includes a “Store & Forward” function which stores videos when a transmission network is inoperable or constrained, and automatically transmits when network availability recovers. Here, V-CRAMMIT was used to streamline 16 MP4 video files which were selected as described below. The original files and their streamlined versions comprised “Uncompressed” and “Compressed” stimulus sets, respectively.
A set of video sequences was compiled from a library archive of deidentified POCUS scans recorded and retained at Brooke Army Medical Center as part of a prior observational study of pulmonary case treatments. Patient Personal Identification Information links were destroyed after this earlier study, so the current study was considered secondary data review. Scans had been recorded in MP4 format using a Philips Lumify POCUS scanner (Koninklijke Philips NV, Amsterdam, Netherlands). Video sequences ranged from 3–8 s in duration and were recorded from various thoracic zones which differed in their frontal, lateral, and vertical location. Sequences were selected from this archive for assignment to four pulmonary pathology conditions: three that were judged to present distinctive pathology indicators—“Pneumonia with Consolidation” (PNA/C), “Pneumothorax” (PTX), “Pleural Effusion” (PE)—and one, “No Finding/No Pathology” (NF/NP), included to make clinicians consider the absence of pathology indicators in the scan. Each video was selected to be an exemplar of indicators for one of the pathologies (or their absence in “NF/NP” videos) and was cross-referenced with the original case’s accompanying CT and patient release diagnosis (also deidentified) as a criterion to confirm accuracy of the condition assignment and avoid situations where several pathologies were present.
Four different POCUS recordings were selected to represent each pathology condition, forming a stimulus set of 16 videos in Uncompressed format. These videos were postprocessed using V-CRAMMIT to generate a matching set of 16 Compressed sequences. V-CRAMMIT yielded reductions in file size ranging from 76–90% (mean = 83%). The complete set of 32 recordings including both Uncompressed and Compressed versions were intermixed and presented in random order in the test session, with each subject reviewing both versions of each recording. Random presentation was included to negate potential familiarity effects; subjects were just as likely to see the Uncompressed vs. the Compressed version of each patient scan first. The stimulus set was double-blind: no cues regarding the video format were available in any trial to either the subject or the investigator.
The study followed a repeated-measures design with two independent within-subject factors: Video format (Uncompressed vs. Compressed) and Pathology condition (PNA/C vs. PTX vs. PE vs. NF/NP, with each possible combination of factors presented in four different patient scans. The design offered a direct comparison of performance across each matched pair of sequences viewed in the two Video conditions.
The task was presented on a laptop (Dell, Round Rock, TX) using a mouse-actuated dashboard interface displayed on a 1920 × 1200, 60-Hz flat-panel screen (Sceptre, City of Industry, CA). The top portion of the dashboard was occupied by a display area where POCUS videos were presented (Fig. 1). The display panel was horizontal (letterbox) in format and occupied the width of the screen. Below the display area, a Play/Replay button was placed for initiating playback. In the lower left portion of the dashboard, a schematic anatomical diagram of the patient’s torso was displayed to indicate where the scan had been performed, i.e., which thoracic zone (among zones 1–8) was being shown in the trial. Within this diagram, the scan zone was highlighted with a red circle. Below the Play/Replay button, five response buttons were placed. These included alternatives naming the four pathology conditions and a fifth button labeled “Inconclusive/Unsure” for the subject to indicate lack of clarity or confidence in any of the response choices. To minimize the possibility of anchoring or other positional biases, the response buttons’ relative positions were shuffled randomly between trials.
Citation: Aerospace Medicine and Human Performance 96, 11; 10.3357/AMHP.6687.2025
Procedure
Subjects were informed that they would review pulmonary scans obtained from various anatomic locations, across a range of magnifications. Instructions named the four possible conditions that might be displayed in each trial, including NF/NP. Instructions stated that the scans would show a variety of appearances, but some trials might resemble one another since the same location could be used for multiple scans.
Upon initiation of the session with the Play/Replay button, the video played the first trial sequence once, after which the subject used the mouse to select the most appropriate response button. Subjects were instructed to treat each trial as an independent evaluation and respond as accurately and swiftly as possible, emphasizing accuracy. Subjects could use the Replay function as many times as desired. Once a response was entered, the next trial began. Accuracy in each trial was recorded as correct or incorrect. “Inconclusive/Unsure” responses were considered incorrect. Response time (RT) was recorded from the onset of the video until the selection of a button. Number of Replays was also recorded.
Statistical Analysis
Performance was assessed using Percent Correct, RT, Number of Replays, and frequency of “Inconclusive/Unsure” responses. Dependent metrics were analyzed using JASP Version 0.19.0 (University of Amsterdam, Netherlands). Two-way repeated-measures ANOVA was performed incorporating Video format and Pathology condition as independent factors with two and four levels, respectively. Where main effects emerged from the multilevel Pathology factor, post hoc contrasts were used using Bonferroni correction for multiple comparisons. Greenhouse-Geisser adjustment was applied to degrees of freedom when the Mauchly test identified departures from sphericity.
RESULTS
Video format did not influence Percent Correct significantly [F(1, 19) = 2.27, P = 0.148]. Pathology influenced Percent Correct [F(3, 57) = 25.99, P < 0.001] without significant interaction. The highest accuracy was recorded in PE trials, followed by PTX, NF/NP, and PNA/C. Post hoc contrasts identified more accurate designations in PE cases than in all other Pathologies (P < 0.005). PNA/C cases were identified less accurately than the other three Pathology conditions (P < 0.001), while accuracy did not differ significantly between PTX and NF/NP (Table I and Fig. 2).
Citation: Aerospace Medicine and Human Performance 96, 11; 10.3357/AMHP.6687.2025
Video format did not influence RT significantly [F(1, 19) = 0.003, P = 0.956]. Pathology influenced RT [F(3, 57) = 16.84, P < 0.001] without significant interaction. The longest and shortest mean RTs were recorded in NF/NP and PE trials, respectively. Contrasts identified shorter RTs in PE than in the other three Pathology Conditions (P < 0.001), with no significant differences among PTX, NF/NP, and PNA/C.
Video format did not influence Number of Replays significantly [F(1, 19) = 0.037, P = 0.850]. Pathology influenced Number of Replays [F(3, 57) = 14.16, P < 0.001] without significant interaction. Subjects used Replay less in PE than in other Pathology conditions (P < 0.005), with no significant differences among PTX, NF/NP, and PNA/C.
Video format did not influence number of “Inconclusive/Unsure” responses significantly [F(1, 19) = 3.065, P = 0.096]. Pathology influenced “Inconclusive/Unsure” responses [F(3, 57) = 3.9, P = 0.013] without significant interaction. Subjects responded “Inconclusive/Unsure” least frequently in PE, with contrasts identifying the difference with NF/NP as significant (P = 0.004).
DISCUSSION
The study compared subjects’ designations of lung pathology reviewed using POCUS scans recorded with vs. without video compression. Percent Correct and RT means did not indicate significantly lower accuracy or delayed decisions when using Compressed vs. Uncompressed video. The absence of main effects for Replay and “Inconclusive/Unsure” metrics indicates that subjects were not discernibly less confident reviewing trials displaying Compressed videos either. These findings were observed without significant interaction from Pathology condition, suggesting that video streamlining did not precipitate additional error vulnerability that was specific to any single pathology. Findings are consistent with the hypothesis that, in this context, clinicians designated pathologies as accurately, swiftly, and confidently using streamlined vs. uncompressed video.
The ability to transmit telehealth images efficiently from remote locales, from aerospace or deep space environments, or in deployed austere situations such as military operations or disaster relief, conveys immense potential benefit. This capability expands access to professional medical expertise and thereby improves quality of care in such locales. In addition, the findings indicate a potential security benefit for deployed military medical personnel whose electromagnetic emissions could be exploited by hostile signals intelligence: increased bandwidth efficiency reduces and shortens targeting vulnerability.
The Pathology condition manipulation identified PE as the condition yielding best performance in this environment. With Percent Correct—the most important performance metric since it assesses clinical accuracy—post hoc contrasts suggested that, in this viewing environment, the visual indicators of excessive pleural fluid were particularly salient and unambiguous compared with other conditions. Conversely, the less accurate performance obtained in PNA/C trials indicates that observable markers for this pathology were less salient and reliable here. RT, Replay, and “Inconclusive/Unsure” metrics also indicated that PE trials yielded the fastest, most confident designations.
This paradigm was designed to evaluate the clinical utility of a video algorithm and was not intended to be comparable to a comprehensive medical evaluation in its diagnostic accuracy. Clinical lung ultrasound studies have yielded sensitivity values of 86% for pneumothorax9 and 93–94% for pneumonia and pleural effusion,10,11 which are higher than the Percent Correct means obtained in three of our four Pathology conditions. Medical diagnosis is complex and requires more tools than a review of a single scan lasting a few seconds, so it is not surprising that overall accuracy fell below 100%, especially considering that in each trial, subjects must make a speeded multiple-choice categorization without clinical context information. While the review of recorded scans emulated some aspects of an asynchronous medical communique, this passive viewing paradigm limited decision-making since it denied subjects the opportunity to direct the scan. The only aspect of review subjects could control was number of replays, which limited the visibility of markers to what was “canned” in the recording. The solution for this is to afford clinicians the opportunity to request or generate their own exploratory scans and, for this purpose, an enhanced version of V-CRAMMIT is under development to afford continuous transmission of streamlined video for telemedicine in real time.
We note that the lower accuracy in PNA/C trials contrasts with relatively high diagnostic sensitivities recorded in clinical studies.11,12 It is possible that the scans presented here represented nonoptimal locations for viewing pneumonia markers, or that these markers were particularly vulnerable to the impediment of passive viewing.
A potential limitation of the study concerns the repeated presentation of each case in order to include both formats. This design was used to enable direct comparison of subjects’ evaluation of stimuli varying only in Video format, at the cost of potential learning effects, which might have contributed data noise. This was addressed by instructing subjects that they would see multiple scans from the same anatomical zone and by using random presentation order to balance potential learning effects between formats.
While it is not claimed that the lack of significantly reduced accuracy and speed in Compressed trials demonstrates unequivocally that streamlined video yields performance equal to uncompressed video, these initial findings are encouraging and motivate further validation of the V-CRAMMIT algorithm. Follow-on studies should confirm its effectiveness in a larger sample of clinicians and include more rigorous screening for medical specialization. (Although deidentified, self-report-based screening reduced recruiting stress, it might have increased variability in subjects’ ultrasound proficiency.) Future investigation should use a variety of image types, including more data-dense formats such as DICOM, where V-CRAMMIT could yield pre-to-post processing file size gains up to 50:1 (compared to the 4:1 ratio obtained here with MP4 files, which already incorporate some compression and might not be clinicians’ preferred format). The next planned validation phase will incorporate the streaming application, currently in development, to afford video telepresence in real time. The final empirical goal of this technical development program will be to compare the effectiveness of clinical diagnoses performed using uncompressed vs. compressed video formats in an environment where data bandwidth is explicitly constrained. In this case, streamlined video could offer the potential not merely to match the performance of uncompressed video, but to surpass it.
Schematic view of interface dashboard including POCUS display panel, Play/Replay button, anatomical scan diagram (note red highlighting indicating that the scan shows Zone 8), and five response buttons. This view shows only a single frame and hence does not show the dynamic information available in the animated video of the test stimuli.
Mean values for A) percent correct, B) reaction times, C) number of replays, and D) frequency of “Unsure/Inconclusive” responses, separated by video format and pathology condition. Error bars represent standard deviations. “PE” = pleural effusion; “PTX” = pneumothorax; “PNA/C” = pneumonia with consolidation; “NF/NP” = no finding/no pathology.
Contributor Notes
