When 3-month-old Mateo was diagnosed with tetralogy of Fallot at a pediatric hospital in Madrid, his parents were handed a whirlwind of technical terms and uncertain timelines. But this time, his cardiologist had a new ally—not to replace judgment, but to sharpen it. Enter DynaTOF, an AI framework quietly transforming how doctors assess one of the most common serious congenital heart defects in children. Born from collaboration across six medical centers and published in eBioMedicine in 2026, DynaTOF doesn’t just analyze echocardiograms—it learns from them, offering a more consistent, data-rich path through the emotional and clinical maze of pediatric heart care.
Tetralogy of Fallot affects about 1 in every 2,000 live births, making it the most prevalent cyanotic congenital heart defect. For decades, echocardiograms have been the gold standard for diagnosis and monitoring, but interpreting these dynamic, grayscale videos demands expertise and time—resources often in short supply. Variability between clinicians can lead to inconsistent measurements, potentially affecting surgical planning and follow-up. DynaTOF tackles this challenge head-on by automating two critical steps: view classification and diameter localization. In testing, the system achieved 94% accuracy in identifying standard echocardiographic views—like apical four-chamber or parasternal long-axis—ensuring the AI is always analyzing the right image. More impressively, it localized and measured key cardiac structures with a mean error of just 0.3 millimeters, a precision that rivals expert sonographers.
But DynaTOF goes beyond measurement. Its real innovation lies in combining visual features from video clips with quantitative data to create a multimodal assessment. In head-to-head comparisons, this dual approach outperformed models using either data type alone by 18%, mirroring the way human clinicians synthesize information. Perhaps most impactful is its ability to predict postoperative recovery. Using pre-surgery echocardiograms, the type of repair performed, and follow-up timing, DynaTOF generates probabilistic recovery trajectories for each child. It won’t tell parents exactly what will happen—but it can flag which patients may need closer monitoring, helping teams prioritize high-risk cases before complications arise.
The framework was trained on a diverse dataset of over 1,200 echocardiographic videos, including healthy controls and conditions that mimic TOF, ensuring it performs well in real-world diagnostic complexity. This isn’t AI in a vacuum; it’s AI built for the messiness of clinical life. As Dr. Yingshuang Gao, lead developer of the framework, puts it: “The most useful AI systems should support clinicians by organizing information, reducing avoidable variation, and highlighting patterns that may otherwise be difficult to see.” With trials now expanding to pediatric centers in Canada and South Korea, DynaTOF represents not just a technological leap, but a quiet shift toward more equitable, consistent care for children born with heart defects—where every beat counts, and every insight matters.