A routine 12-lead electrocardiogram may contain more information about the heart’s anatomy and performance than clinicians can extract by visual inspection alone, according to a new study by J.A. Brody, V. Yogeswaran, K.L. Wiggins and colleagues published in Nature Communications. The research focuses on the left atrium, a chamber crucial to blood flow that is usually assessed with imaging rather than the inexpensive, widely available ECG. Echocardiography, cardiac magnetic resonance imaging and computed tomography can reveal chamber size and mechanical behavior, but these tests require specialized equipment, trained personnel and, in some cases, substantial cost or patient preparation. The left atrium receives oxygen-rich blood from the lungs and transfers it to the left ventricle. Changes in the atrium can develop gradually in response to high blood pressure, valve disease, heart failure and abnormal rhythms such as atrial fibrillation. A conventional ECG records voltage changes across the skin through 12 leads positioned on the limbs and chest. The study’s central premise is that subtle patterns in the waveform may encode indirect clues about left atrial size and function that are too complex for human observers to recognize consistently. Deep learning uses multilayered artificial neural networks that transform raw or processed ECG signals into increasingly abstract representations, allowing the model to learn statistical relationships between electrical activity and cardiac mechanics. Such a system could turn an ECG into predictions of characteristics normally obtained through imaging, including structural measures like dimensions or volume and functional measures describing contraction, relaxation or contribution to ventricular filling.