In January 2026, the FDA expanded clearance for an AI platform that talks non-sonographers through acquiring diagnostic-quality ultrasound images. In May, a systematic review and meta-analysis in BMC Emergency Medicine pooled seven studies and reported that convolutional neural networks detect free intraperitoneal fluid on FAST images with a sensitivity of 91.1% and a specificity of 97.5%. In July, a group at the Johns Hopkins Applied Physics Laboratory published a mixed-reality system that projects a morphable anatomical atlas - built from 180 CT scans - onto the patient’s torso through a headset, showing an untrained operator exactly where to place the probe.
The direction of travel is unmistakable. The FAST examination, the most operator-dependent investigation in the trauma bay, is being systematically de-skilled; within a few years the machine will acquire the images, read them, and hand you an answer. It is worth pausing to ask what question that answer is to. Because at the centre of this enthusiasm sits an uncomfortable fact: across three decades of use, and every randomised trial conducted in that time, nobody has demonstrated that performing a FAST scan improves a single patient-important outcome. Not mortality. Not length of stay. Not missed injuries. We are about to automate a test we have never been able to show works.
A test that answered the right question in 1995
FAST earned its place honestly, displacing diagnostic peritoneal lavage - invasive, with a meaningful complication rate and a habit of triggering non-therapeutic laparotomies - with something non-invasive, repeatable and quick. ATLS adopted it as an adjunct to the circulation element of the primary survey, where it has remained through every subsequent edition, including ATLS 11, launched in September 2025 with its headline shift to the xABCDE sequence.
The logic was sound in the imaging environment of the mid-1990s, when CT meant transferring an unstable patient to a distant scanner and a delay measured in tens of minutes. A bedside test that could tell you within ninety seconds whether there was blood in the abdomen genuinely changed what happened next. That environment no longer exists in most systems where ATLS is taught. The modern resuscitation room frequently contains a CT scanner or sits immediately adjacent to one, and whole-body CT of a polytrauma patient takes minutes from arrival. The gap FAST was invented to fill has largely closed. The test has not moved.
What the FAST detects - and what it structurally cannot
The most comprehensive accuracy data come from Netherton and colleagues in the Canadian Journal of Emergency Medicine, pooling 75 studies and 24,350 patients. For intra-abdominal free fluid, sensitivity was 74% (95% CI 72.6–75.8) with specificity of 98%. For pneumothorax, sensitivity was 69% (95% CI 66.0–72.7) with specificity of 99%. For pericardial effusion, sensitivity was 91% and specificity 94%.
Read those as a clinician rather than a statistician. A sensitivity of 74% means roughly one in four patients with free intraperitoneal fluid has a FAST that does not show it; for pneumothorax it is nearly one in three. These are not the characteristics of a rule-out test, and the authors said so plainly: eFAST is useful for ruling in, and their data do not support its use for ruling out.
Those pooled figures are also weighted towards experienced operators in academic centres. A prospective study in BMC Emergency Medicine in November 2025 examined eFAST performed by newly trained emergency physicians - the group most ATLS candidates actually belong to. Specificity was 100% across every application. Sensitivity for haemoperitoneum was 81%, for pneumothorax 59%, and for haemothorax 36%, with an area under the curve of 0.68 - barely better than a coin weighted slightly in your favour.
Then add the anatomy, which no amount of training or computation will change. FAST does not see the retroperitoneum, where the duodenum, pancreas, kidneys and great vessels live. It does not see hollow viscus injury until perforation has produced enough fluid to pool, nor diaphragmatic injury, nor solid organ injury that has not yet bled into the peritoneum - which describes a substantial proportion of splenic and hepatic lacerations at the moment the probe is applied.
The outcome problem
Diagnostic accuracy is not the same as clinical usefulness, and it is on usefulness that FAST has never delivered. The Cochrane review of ultrasound-based algorithms in blunt abdominal trauma identified four randomised trials. Pooled mortality across three of them, comprising 1,254 patients, gave a relative risk of 1.00 (95% CI 0.50–2.00). FAST-based pathways did reduce CT use - risk difference −0.52 (95% CI −0.83 to −0.21) - but the reviewers refused to read this as a benefit: fewer CT scans achieved by a test with FAST’s sensitivity may simply mean more missed injuries. They rated the evidence poor quality.
The single best trial remains Holmes and colleagues in JAMA in 2017. Nine hundred and twenty-five haemodynamically stable children with blunt torso trauma at a level I centre were randomised to FAST plus standard evaluation, or standard evaluation alone. CT was performed in 52.4% of the FAST arm and 54.6% of controls. Mean emergency department length of stay was 6.03 hours versus 6.07. Median charges were $46,415 versus $47,759. Missed intra-abdominal injuries: one in the FAST arm, none in the control arm. Nothing moved.
The honest summary is that FAST is a test clinicians find reassuring, which adds about ninety seconds to a resuscitation, and which has never been shown to alter what happens to the patient.
The British complication
UK candidates should read this section twice, because national guidance and course doctrine are not merely different here. They are opposed.
NICE guideline NG39 on major trauma states at recommendation 1.5.32: do not use FAST or other diagnostic imaging before immediate CT in patients with major trauma. At 1.5.33: do not use FAST as a screening modality to determine the need for CT in patients with major trauma. Recommendation 1.5.30 warns that a negative FAST does not exclude intraperitoneal or retroperitoneal haemorrhage, and 1.5.29 confines FAST to the minimum imaging needed to direct intervention in the haemodynamically unstable patient who is unresponsive to resuscitation. Prehospitally, recommendation 1.3.2 permits eFAST only where a specialist team with ultrasound is immediately available and onward transfer will not be delayed, with 1.3.3 adding that a negative eFAST of the chest does not exclude a pneumothorax.
So in England and Wales, FAST has one narrow surviving indication- the patient too unstable to reach the scanner - and is actively discouraged everywhere else. ATLS continues to teach it as a routine adjunct to the primary survey.
Both positions are defensible. ATLS is a global course that must work where the nearest CT scanner is two hundred kilometres away, and ATLS 11 makes a virtue of this in its new language of “standardised flexibility”. NICE is writing for a mature, geographically dense network with CT in the resus room. But the candidate sitting an ATLS examination while working in a UK major trauma centre is asked to hold two incompatible doctrines at once, and the course does not flag the tension.
What the AI actually fixes
Against that background, consider what the 2026 meta-analysis by Çelik and colleagues shows. Seven studies, 2,332 patients, more than 34,000 images and video clips. Pooled trauma-only sensitivity 91.1% (95% CI 77.9–96.8), specificity 97.5% (95% CI 95.3–98.7), area under the curve 0.98, heterogeneity below 1%; in the right upper quadrant view, sensitivity reached 93.4%. Taken at face value, that is a machine beating the pooled human sensitivity of 74% by a wide margin.
The caveats are severe, and to their credit the authors state them without hedging. All seven studies were retrospective. Only one performed external validation; the rest relied on internal hold-out sets and cross-validation, which the authors concede “may inflate performance estimates”. Five of seven were rated high or unclear risk of bias for patient selection under QUADAS-AI. Not one reported dedicated accuracy data for the pelvic view - the view most likely to turn positive first in a supine patient - and evidence for pericardial effusion came from a single study. Most importantly, none evaluated real-time integration into an emergency department or prehospital workflow. These algorithms were fed pre-acquired, curated images, and the authors are unambiguous that such performance “does not necessarily translate into improved acquisition quality, faster decision-making, or better patient outcomes in real-world trauma care”.
And acquisition, not interpretation, is where inexperienced operators fail. The Johns Hopkins mixed-reality study is the most honest data point in this literature. With holographic anatomical guidance overlaid on the patient, novice operators improved the proportion of eFAST examinations that were diagnostically sufficient from 13% to 25%. That is a doubling, and a genuine achievement. It also means three-quarters of scans performed by guided novices remained uninterpretable. A 91%-sensitive algorithm is worth nothing applied to an image that does not contain the anatomy.
The bottleneck was never perceptual
Here is the argument I would put to anyone excited by AI-assisted FAST. The failure mode of FAST in contemporary trauma care is not that clinicians cannot recognise free fluid when it is on the screen. It is that the test answers a question which, in most patients, does not change the decision.
Take the two ends of the spectrum. The hypotensive patient with a distended abdomen who is not responding to resuscitation is going to theatre: a positive FAST confirms your intention, and a negative FAST at 74% sensitivity cannot safely dissuade you. The stable patient with a seatbelt sign is going to CT, because CT identifies the retroperitoneal, hollow viscus and non-bleeding solid organ injuries that FAST structurally cannot see. A positive FAST does not change the destination; a negative one must not. The population in which FAST genuinely arbitrates - unstable enough that CT is unsafe, but stable enough that the decision to operate is in real doubt - exists, but it is small. Perfecting image interpretation makes the test better at what it was already good at and leaves untouched what it is bad at.
There is a second-order risk worth naming. A confident automated output changes how a negative result feels. A clinician who knows their own scan was rushed treats a negative result with appropriate suspicion; a clinician handed “no free fluid detected” by a cleared medical device is being invited into precisely the false reassurance the sensitivity data forbid. Automation bias should be expected to bite hardest where a test’s negative predictive value is weakest. FAST is such a test.
The strongest case on the other side
That case deserves to be put at full strength, because it is better than sceptics usually allow.
Specificity is the point, and it is excellent. Ninety-eight per cent for free fluid, 99% for pneumothorax. In a crashing patient a positive FAST is close to diagnostic and arrives in ninety seconds; few tests buy that much certainty that quickly. Properly understood, the 74% sensitivity is not an indictment of the test but an instruction about how to use the result. And 91% sensitivity for pericardial effusion - a diagnosis otherwise reached largely by inference, whose treatment is immediate and dramatic - would on its own justify keeping the subxiphoid view in the primary survey.
The argument from geography is the strongest of all. NICE writes for a system with CT in the resus room; most of the world has no such luxury. In prehospital care, in military and austere practice, and across most low- and middle-income trauma systems, FAST is not competing with CT - it is competing with nothing. The June 2026 systematic review in the Journal of Ultrasonography, covering 110 studies, reports prehospital eFAST specificity of 98% and paramedic interpretation accuracy of 97.35%. It is precisely where operators are least experienced and the alternative is no imaging at all that AI acquisition guidance has most to offer. An algorithm that lifts a rural clinician’s scan from uninterpretable to diagnostic is not solving a first-world problem.
The evidence is thin rather than negative. The Cochrane review found four small, poor-quality trials, and a mortality risk ratio whose confidence interval spans 0.50 to 2.00 is compatible with halving mortality. Writing after the Holmes trial, Moore and Liu put it neatly: there was “no evidence of harm, just no evidence of help”. They added a point that has aged well - that abandoning FAST in stable patients erodes the proficiency needed to perform it competently in unstable ones, a self-fulfilling prophecy in which the test deteriorates because we stopped practising it.
Serial scanning may be the answer the field has missed. The same 2026 review reports repeated eFAST in stable blunt thoracoabdominal trauma achieving 100% sensitivity and 98.7% specificity. If the problem is a single snapshot taken before enough blood has accumulated, the remedy may be more scans rather than fewer - and automated acquisition makes repeat scanning cheap.
The definitive trial is running now. Holmes and colleagues published a protocol in Trials in December 2025 for a multicentre randomised trial of 3,194 to 4,346 haemodynamically stable children across six sites, powered on CT utilisation within 24 hours and on missed or delayed intra-abdominal injury, completing in May 2027. That is the trial paediatric FAST has needed for twenty years, and judgement should probably wait for it.
Where this leaves the trauma bay
Keep the probe; shrink the claim. FAST is a rule-in test for the unstable patient and a triage instrument where CT is unavailable. It should not determine whether a stable major trauma patient goes to CT, and a negative result should never be charted as reassurance. AI that improves image acquisition in inexperienced hands is a real contribution to prehospital, military and resource-limited care, and belongs there. AI that improves interpretation in a resuscitation room with a CT scanner forty feet away is solving a problem that stopped mattering some time ago. What deserves resistance is the quiet upgrade in confidence that automation brings - treating a machine-read negative FAST as more informative than a human-read one, when the limits of the test are anatomical rather than perceptual, and no algorithm can see around them.
What this means for your ATLS exam
• FAST and eFAST remain adjuncts to the circulation element of the primary survey in ATLS 11. Whatever the evidence debate, answer according to course doctrine. ATLS 11’s headline change is the xABCDE sequence, not a revision of imaging adjuncts.
• Positive FAST plus haemodynamic instability equals laparotomy. This is the single most heavily examined FAST fact. Do not divert such a patient to CT.
• A negative FAST never excludes intra-abdominal injury. Pooled sensitivity for free fluid is approximately 74%. Examiners test this directly, often by offering a negative FAST as a distractor in a patient with a concerning mechanism.
• Know what FAST cannot see: the retroperitoneum, hollow viscus injury, diaphragmatic injury, and solid organ injury that has not bled into the peritoneum. These are the classic stems.
• Know the views. FAST: pericardial (subxiphoid), right upper quadrant (Morison’s pouch), left upper quadrant (splenorenal recess), and pelvis (rectovesical pouch or pouch of Douglas). eFAST adds bilateral anterior chest views for pneumothorax.
• Carry the figures. Free fluid 74% sensitive, 98% specific; pneumothorax 69% sensitive, 99% specific; pericardial effusion 91% sensitive, 94% specific. Quoting them separates a good viva from an adequate one.
• UK candidates: hold both positions deliberately. Answer ATLS doctrine in the examination; practise to NICE NG39 on the shop floor, where FAST is not to be used before immediate CT or as a screening test for CT. Being able to articulate why the two differ — resource context, not disagreement about physiology — is a strong viva answer.
• AI-assisted FAST will not be examined. It appears in no current guideline and in no edition of ATLS. But a candidate who can explain why a more accurate reading of an unproven test is not the same as a better test is demonstrating exactly the critical appraisal that marks out a strong performance.
References
1. Netherton S, Milenkovic V, Taylor M, Davis PJ. Diagnostic accuracy of eFAST in the trauma patient: a systematic review and meta-analysis. CJEM. 2019;21(6):727–738. doi:10.1017/cem.2019.381. https://www.cambridge.org/core/journals/canadian-journal-of-emergency-medicine/article/diagnostic-accuracy-of-efast-in-the-trauma-patient-a-systematic-review-and-metaanalysis/06E27D34E443D284F3ACAA4F6B6BA3F3
2. Stengel D, Rademacher G, Ekkernkamp A, Güthoff C, Mutze S. Emergency ultrasound-based algorithms for diagnosing blunt abdominal trauma. Cochrane Database of Systematic Reviews. 2015;(9):CD004446. doi:10.1002/14651858.CD004446.pub4. https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD004446.pub4/full
3. Holmes JF, Kelley KM, Wootton-Gorges SL, et al. Effect of abdominal ultrasound on clinical care, outcomes, and resource use among children with blunt torso trauma: a randomized clinical trial. JAMA. 2017;317(22):2290–2296. doi:10.1001/jama.2017.6322. https://jamanetwork.com/journals/jama/fullarticle/2631528
4. Çelik A, Topaloğlu E, Yazıcı MM. Diagnostic accuracy of AI-assisted point-of-care ultrasound for abdominal free fluid detection in FAST trauma assessment: a systematic review and meta-analysis. BMC Emergency Medicine. 2026;26:191. doi:10.1186/s12873-026-01616-6. https://link.springer.com/article/10.1186/s12873-026-01616-6
5. Armiger RS, Knight AE, Carneal CM, Kuo NP, Perrino KM, Deanehan JK. Development and evaluation of an anatomically guided system for ultrasound image acquisition in trauma assessment. Journal of Medical Imaging (Bellingham). 2026;13(4):045001. doi:10.1117/1.JMI.13.4.045001. https://pubmed.ncbi.nlm.nih.gov/42689164/
6. Jusoh AF, Yahaya R, Fauzi MH, Abdull Wahab SF. Point-of-care ultrasound in trauma: a systematic review of recent literature. Journal of Ultrasonography. 2026;26:18. doi:10.15557/JoU.2026.0018. https://jultrason.pl/assets/pdf/artykuly/8-jou-00114-2025-jusoh-hr-pdf.pdf
7. Buyurgan ÇS, Yarkac A, Bozkurt S, et al. Diagnostic accuracy of E-FAST examination performed by newly trained emergency physicians and its impact on clinical outcomes. BMC Emergency Medicine. 2025;25:226. doi:10.1186/s12873-025-01386-7. https://link.springer.com/article/10.1186/s12873-025-01386-7
8. National Institute for Health and Care Excellence. Major trauma: assessment and initial management. NICE guideline NG39. Recommendations 1.3.2, 1.3.3, 1.5.29, 1.5.30, 1.5.32, 1.5.33. https://www.nice.org.uk/guidance/ng39/chapter/recommendations
9. Holmes JF, Tancredi DJ, Kelley KM, et al. Abdominal ultrasound (FAST) in hemodynamically stable children with blunt abdominal trauma: study protocol for a randomized controlled trial. Trials. 2025;26:564. doi:10.1186/s13063-025-09137-6. https://link.springer.com/article/10.1186/s13063-025-09137-6
10. Moore C, Liu R. Not so FAST - let’s not abandon the pediatric focused assessment with sonography in trauma yet. Journal of Thoracic Disease. 2018;10(1):1–3. doi:10.21037/jtd.2017.12.37. https://jtd.amegroups.org/article/view/18180/14633
11. American College of Surgeons. Trauma care gets major upgrade with launch of ATLS 11. ACS Brief, 16 September 2025. https://www.facs.org/for-medical-professionals/news-publications/news-and-articles/acs-brief/september-16-2025-issue/trauma-care-gets-major-upgrade-with-launch-of-atls-11/
12. Diagnostic Imaging. FDA issues expanded clearance for UltraSight’s cardiac ultrasound echo stewardship program. January 2026. https://www.diagnosticimaging.com/view/fda-expanded-clearance-ultrasight-cardiac-ultrasound-echo-stewardship-program