Adaptive entropy constrained transform coding of magnetic resonance image sequences

Compression of magnetic resonance images (MRI) has proved to be more difficult than other medical imaging modalities, and attempts at utilizing interslice dependencies for more efficient coding have so far met with little success. On the other hand, the increasing amounts of MRI data generated every day in hospitals makes this particular data compression problem very important. In this paper, we present an adaptive, entropy constrained transform coder that also employs a new interslice estimator. Previous attempts at interslice coding of MRI have all used a piecewise uniform, discontinuous translational model. We propose a continuous piecewise affine model for interslice dependencies, whose implementation is performed through a triangle-based matching (TBM) algorithm. The residue frames from the interslice estimator are coded through an entropy constrained quantizer, applied to the block discrete cosine transformed (DCT) residue frame. Given that image statistics vary spatially, the optimal spectral distribution of bits is nonuniform. We address this issue through dividing the image into several “activity” regions. Each activity class will have a separate bank of entropy constrained scalar quantizers (ECSQs) tailored to its particular average statistics. The proposed coder was used to encode a sequence of a human heart. An improvement of almost 1 db over nonadaptive interslice and 0.3 dB over adaptive intraslice coding was obtained