All ultrasound machines assume sound travels at 1540 m/s. When you're imaging through fat (1450 m/s), the machine slightly miscalculates depth — creating the "speed artifact" that shifts structures deeper than they are.
Wavelength λ = c/f. A 15 MHz probe in soft tissue has λ ≈ 0.1 mm — near-capillary resolution. But attenuation scales with frequency: 15 MHz loses signal 30× faster per cm than 0.5 MHz.
You shout. You hear echoes from different walls at different times. The time delay tells you distance. The loudness tells you the reflectivity of the surface. Ultrasound is exactly this — just at 7 million Hz.
Air has an acoustic impedance 4000× lower than tissue. Nearly 100% of the wave reflects at any air-tissue interface. Bone creates the opposite problem — nearly total reflection and extreme attenuation of what passes through.
Jean le Rond d'Alembert formulated the wave equation in 1747. Siméon Poisson extended it to 3D. These were pure mathematics — the medical application came 200 years later when Ian Donald first used ultrasound clinically in obstetrics (1950s). The physics waited for the technology to catch up.
Most AI ignores the physics it's built on. Intracav can incorporate speed-of-sound constraints and attenuation models directly into the inference layer — giving the system anatomical priors that pure pixel-based models lack.
Attenuation model → TGC guidance
Reflection coefficients → boundary detection
Tissue priors → vessel vs non-vessel