An AM-FM video image representation is developed and applied to M-mode ultrasound. The AM-FM representation describes the video in terms of an AM-FM series expansion. For estimation, the fundamental AM-FM harmonic is estimated using a collection of bandpass filters that are adaptively selected so that one with the maximum response is selected. For each bandpass filter, the output image is downsampled depending on the spectral support of the filter. Using results from multidimensional Frequency Modulation theory, it is shown that a fast separable implementation is possible. In turn, good estimates are obtained from a collection of only two separable, bandpass filters approximated with just a small number of 11 coefficients and the recently developed SIMD-FFT. The fundamental AM-FM harmonic is seen to capture the essential structure of the image and simple thresholding of the amplitude estimate yields very good segmentation results.