Author + information
- Guglielmo M. Trovato, MD∗ ( and )
- Marco Sperandeo, MD
- ↵∗Department of Medical and Pediatric Science, The University Hospital of Catania, Via Santa Sofia 78, 95100 Catania, Italy
Raso et al. (1) describe a software claimed as “a robust approach for developing a portable device for the individualized and automatic detection of pulmonary interstitial edema or fibrosis,” designed to detect B-lines. We are concerned that these are merely artifacts without anatomic correspondence (2), which cannot reasonably or reliably be considered markers specific for any disease, pulmonary interstitial edema or fibrosis or otherwise. These findings are seen in normal subjects, and even in pneumonectomized patients, that is, where there is no lung present (2). The ultrasound artifacts described in the preceding text are particularly frequent when low-frequency phased-array probes are used (1).
Moreover, in the light of the Razo and colleagues (1) appropriate acknowledgement that B-lines cannot be quantified exactly, we wonder whether a soft computing-based B-line analysis through a knowledge-based model for an objective, operator-independent, automated, and quantitative classification of the severity of pulmonary interstitial syndrome will amplify, or minimize, this a priori uncertainty of measurements. Despite the claim that the model correctly identified normal subjects in 100% of cases and was able to discriminate the 3 levels of severity of pulmonary interstitial edema and pulmonary fibrosis, apparently using the B-line count alone, we respectfully ask how the tool would perform in other conditions, because the B-line count is also notoriously operator- and setting-dependent (2), and whether this procedure, acknowledged as nonspecific (1), is able to detect and discriminate different superimposed or overlapping conditions of similar appearance (2).
The description of laser-like vertical hyperechoic reverberation artifacts moving synchronously with respiration remind us of the artifacts seen in electrocardiographic tracings of pulmonary edema and several lung diseases. As your readers will be aware, these artifacts are unrelated to electrical activity of the heart, and do not reflect cardiac potentials on the body surface; they merely distort the electrocardiogram and disappear with improvement or death, exactly like B-lines.
Although, as Raso and colleagues (1) rightly claim, an aid to diagnosis of such diseases by such a cost- and time-efficient manner could have a high impact on public health, we note that the gold standard considered in this case was still a physician with more than 9 years’ experience (1). We wonder what objective measures were considered in the clinical diagnoses (high-resolution computed tomography, or pulse oxygen saturation, or even pulmonary auscultation). It is unclear whether an imperfect model that incorrectly classifies 5% of patients, apparently standardized on the basis of the observations of a single expert, without an interobserver variability report, is sufficiently robust. We are indeed “moving toward pervasive health care systems in human-oriented environments,” but this will only be a good thing if artificial intelligence and telemedicine are based on robust, reliable measures, and not on the ghost in the machine.
Please note: The authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- 2015 American College of Cardiology Foundation