Echocardiography pretraining offers no edge over video for lung ultrasound tb screening
Wrong Organ, Right Physics: Transferring Echocardiography Pretraining to Lung Ultrasound for Tuberculosis Screening
Computer Vision and Pattern RecognitionMachine Learning
Summary
Finding early signs of tuberculosis (TB) in lungs using ultrasound images is helpful but data is limited for training AI to do this. The authors tested if using a model trained on heart ultrasounds (echocardiography) helps more than one trained on generic videos because both use similar imaging physics. They found no meaningful difference between the two in identifying TB from lung ultrasounds. Instead, how features from these models are standardized before classification had a bigger impact on performance.
What this means in practice
- •For medical imaging ai developers: Improve TB lung ultrasound classification by focusing on feature standardization rather than the choice between echocardiography or video pretrained encoders.
- •For ultrasound device software teams: Integrate better feature conditioning techniques in lung ultrasound analysis software to enhance tuberculosis screening accuracy without additional pretrained data needs.
Authors
Christiaan M. Geldenhuys, Joshua M. Jansen van Vüren, Véronique Suttels, Trevor Brokowski, Ablo P. Wachinou, Mary-Anne Hartley, Rensu P. Theart, Grant Theron, Thomas R. Niesler
Abstract
Lung ultrasound (LUS) is attractive for tuberculosis (TB) screening at primary-care level, but labelled cohorts are small. Echocardiography carries no such constraint, while sharing the same underlying ultrasound imaging physics, signal processing and B-mode appearance as LUS. We ask whether an encoder pretrained on that high-resource ultrasound domain carries representations that remain usable in the low-resource one. Only the encoder varies, across seventeen encoders spanning three architecture families. Among them, a latent-predictive video encoder pretrained on generic video (V-JEPA2-L) and its echocardiography counterpart (EchoJEPA-L) differ in pretraining corpus alone. The choice among these encoders does not resolve the classification, the whole family spanning 2.50 percentage points against a measurement resolution of 2.71. What moves the task instead is feature conditioning. Standardising the features between the encoder and the classifier improves all seventeen encoders by a mean of +1.23 percentage points at $p=1.5\times10^{-5}$. On the held-out test set every encoder selected on the development folds stands above the baseline system by up to +2.57 percentage points of area under the receiver operating characteristic curve (AUROC), and specificity at 90% sensitivity reaches 79.3% against 60.3%. The contrast specified in advance, EchoJEPA-L against V-JEPA2-L, measures -0.16 percentage points at $p=0.926$. We therefore find no evidence that shared ultrasonic physics alone makes echocardiography a more productive pretraining corpus than generic video, and any advantage, if present, is smaller than this cohort can resolve. The video encoders receive replicated still images, however, so whether this absence of an effect reflects the pretraining domain or a video encoder applied to static frames cannot be separated. The limiting factor is the labelled cohort rather than the encoder.