Chronic liver disease is common and can remain clinically silent while fat accumulates in the liver and fibrosis progresses. Two useful non-invasive measurements are the controlled attenuation parameter (CAP), which reflects hepatic fat, and liver stiffness measurement (LSM), which reflects stiffness associated with fibrosis. FibroScan provides both measurements, but it requires dedicated equipment and is not available in many smaller or resource-constrained centres.
Routine B-mode ultrasound is far more widely available and is already part of standard liver assessment. It contains visible information related to steatosis, including changes in echogenicity and attenuation, but fibrosis is harder to infer from appearance alone. This creates a practical access problem: patients may have a routine ultrasound without access to the specialised test that would provide quantitative CAP and stiffness values. A useful alternative would estimate those measurements from data that clinics already collect.
Clinicians need quantitative estimates they can obtain from routine examinations, on equipment a local centre already has. For liver fat, the estimate should make use of the information already visible in good-quality B-mode ultrasound. For stiffness, it should be able to incorporate routine clinical and laboratory measurements when those add information that the image does not contain. The output should remain a continuous CAP or LSM estimate, and its expected error must be clear enough for clinicians to understand that it is an approximation of FibroScan, not a substitute for histology.
We are developing models that estimate continuous FibroScan CAP and liver stiffness values from routine B-mode ultrasound, using multiple frames per patient and routine clinical and laboratory measurements where they add information the image does not carry. Development uses a retrospective cohort of roughly 1,400 patients with patient-level train, validation, and test separation.
Using clear and partial frames without biomarkers produced the best CAP result, with test R²=0.404 and MAE=31.0 dB/m; adding 30 routine clinical and laboratory variables did not improve performance. In the clear+partial setting, it achieved test R²=0.545 and MAE=2.43 kPa. Each setting was repeated across three seeds. FibroScan is the reference standard for both tasks, so these results measure agreement with FibroScan rather than biopsy-defined histology.