Lumify Diagnostic Ultrasound System

The Philips Lumify Diagnostic Ultrasound System employs an expanded B-lines software feature, utilizing a locked machine learning algorithm on compatible Android devices, to automatically detect, count, and identify merged B-lines in lung ultrasound cine loops for point-of-care assessment of pulmonary conditions. This system aims to enhance diagnostic capabilities by providing real-time, quantitative analysis of B-line artifacts, with clinical studies demonstrating good performance in sensitivity, specificity, and correlation with clinician assessments for merged B-lines and B-line counts respectively.

The Philips Lumify Diagnostic Ultrasound System is a mobile, software-controlled device intended for diagnostic ultrasound imaging across multiple modes (B-mode, Color Doppler, Pulsed Wave Doppler, M-mode) and a wide range of clinical applications including fetal/obstetric, abdominal, cardiac, and lung imaging. A significant enhancement is its expanded B-lines software feature, which operates on Android-compatible mobile devices and utilizes the Philips C5-2, L12-4, or S4-1 transducers for detailed lung examinations. This feature is designed to assist clinicians in assessing lung involvement in conditions such as pneumonia, pulmonary edema, lung contusion, and acute respiratory distress syndrome (ARDS) by identifying B-lines – hyperechoic artifacts resulting from air-water mixture – and specifically, "merged B-lines," which occur when numerous B-lines coalesce, indicating potentially more severe fluid accumulation and making individual counting difficult. The system allows for image acquisition in up to 12 lung zones (8 anterior, 4 posterior). The expanded B-lines feature leverages a locked machine learning algorithm to analyze lung ultrasound cine loops in real-time. This algorithm automatically counts the number of B-lines in each frame, determines if B-lines are merged, and highlights these findings. After acquisition, the software reports an output category for the loop: "0 B-lines", "1-2 B-lines", "3+ B-lines", or the symbol ⊥ if merged B-lines are detected. Clinicians can review the highlighted B-lines and merged B-lines and have the option to accept or adjust the reported category. A clinical performance study using 416 lung ultrasound video loops from 157 subjects (distinct from the training dataset) demonstrated that for merged B-line detection, the algorithm achieved an overall sensitivity of 0.83 (95% CI: 0.77, 0.88) and specificity of 0.92 (95% CI: 0.88, 0.96) compared to clinician majority agreement. For B-line counting, the intraclass correlation coefficient (ICC) was 0.91 (95% CI: 0.89, 0.93), indicating strong agreement with clinician assessments.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the machine learning algorithm differentiate between numerous, closely packed individual B-lines and actual "merged B-lines," especially in borderline cases or varying image quality?; What was the inter-observer variability among clinicians in establishing the "ground truth" for B-line counts and merged B-line presence in the clinical performance study dataset, and how does this compare to the algorithm's consistency?; Could the performance of the merged B-line detection (sensitivity/specificity) be further stratified by the underlying etiology of the B-lines (e.g., cardiogenic pulmonary edema vs. ARDS vs. pneumonia)?; How does the system account for, or guide users to mitigate, potential confounding factors in B-line assessment, such as rib shadows or pleural irregularities not related to air-water mixtures?; What is the typical processing time for the algorithm to analyze a cine loop and provide the B-line count and merged B-line status on various compatible Android devices with differing processing capabilities?