Severe schistosomal fibrosis proved hardest to read, even in a large field cohort
In a population-based cohort of 3302 people in rural Uganda, periportal fibrosis linked to Schistosoma mansoni exposed a basic diagnostic problem: sonographers did not consistently see the same liver pattern when they reviewed the same ultrasound findings. The study used point-of-care ultrasound images and video from the SchistoTrack cohort, then asked second sonographers to read curated liver images while blinded to the original field diagnosis. Even with standardized collection and a defined pattern system, disagreement persisted across the full spectrum of fibrosis.
The sharpest signal came from the severe end of the spectrum. Patterns E and F, which represent advanced disease, had the weakest agreement, with κ values of 0.430 and 0.149, respectively. By contrast, pattern D showed the most agreement, with κ = 0.717. A single score system that treats all fibrosis patterns as equally distinct is already struggling at the bedside.
Adjacent fibrosis categories blurred together in practice
The cohort design matters because this was not a small, referral-based sample selected for obvious disease. Six sonographers examined participants aged 5 to 92 years during 2024, and 32.8% of the cohort had abnormal liver fibrosis patterns from B through F. Among those participants, 2074 images from 1040 individuals were read again by a second sonographer. Broad pattern agreement reached 73.0%, yet 94.0% of second readings still fell within one step of the field diagnosis. In other words, the disagreements were usually close, not random.
That close-range error pattern points toward a continuum rather than a set of clean compartments. Periportal fibrosis does not behave like a binary lesion with a bright line between normal and abnormal. It evolves through gradual tissue remodeling, portal tract thickening, and architectural distortion, and the ultrasound patterns appear to reflect a moving biological process more than a fixed diagnostic label.
Second readers also judged their own confidence, probe position, breathing or movement interference, and fat interference. Those technical factors were associated with disagreement. Participant characteristics, by contrast, were not consistently linked to discordance. The practical implication is straightforward: image quality and scanning conditions shaped interpretation more than age or other baseline traits did.
Video sweeps performed far worse than still images
The most striking failure involved ultrasound video. Among 1054 videos reviewed by second sonographers, agreement was essentially absent: κ = 0.001 for the presence of periportal fibrosis, and individual pattern κ values ranged from 0 to 0.062. Video sweeps, at least in this setting, did not rescue interpretive uncertainty.
That finding has direct consequences for field screening programs. Video may seem attractive because it captures more anatomy and gives the reader more context, but more motion does not automatically translate into more clarity. Without stable probe position, controlled acquisition, and a shared interpretive framework, the extra information may simply amplify noise. In a disease where severe cases need fast triage, a modality that cannot reliably preserve agreement becomes difficult to defend as a substitute for carefully curated still images.
The WHO Niamey framework may need a more continuous model
The study used an adapted and standardized version of the WHO Niamey protocol, with liver fibrosis patterns labeled A through F and periportal fibrosis defined as patterns C through F. The results do not reject the protocol outright, but they do expose its friction points. Adjacent categories were often confused, severe disease was least reproducible, and agreement clustered around the same sonographer more than around the participant.
A continuous severity model may fit the biology better than a rigid pattern ladder. Liver injury caused by schistosomal eggs drives chronic inflammation and fibrotic remodeling along the portal tracts. As fibrosis accumulates, ultrasound appearances likely shift along a spectrum rather than jumping from one discrete state to the next. A grading system built to capture progression, instead of enforcing hard borders, could better match what sonographers are actually seeing.
The data also argue for more than one reader when severe disease is suspected. Severe patterns are the ones most likely to carry clinical weight, because they mark patients at higher risk for complications and may influence referral or treatment priorities. If the hardest cases are also the least reproducible, second reading becomes a safety mechanism rather than an academic luxury.
A large field study, but still a preprint with operational limits
This is the largest diagnostic study of inter-sonographer agreement for schistosomal liver fibrosis patterns in the cohort described here, and the scale gives the findings unusual force. It included a broad community sample, standardized image capture, and blinded re-reading by second sonographers. Even so, the results come from a preprint that has not undergone peer review, and the image data are not openly available because of privacy and consent restrictions tied to the ongoing cohort.
The study also makes clear where implementation could fail. In areas where Schistosoma mansoni remains endemic, ultrasound is often the only realistic imaging tool for detecting periportal fibrosis. If readers disagree most when the disease is most severe, then program design has to account for that weakness. Better acquisition training, clearer quality criteria, and mandatory confirmation of severe patterns are obvious next steps. The unresolved problem is not whether ultrasound can detect fibrosis at all; it is how to make the reading process consistent enough to protect the patients whose livers are already most damaged.