Based on the analysis of the current kinds of Region of Interest (ROI) coding, a mask embedded SPIHT algorithm was proposed, which supported multiple ROI in arbitrary shape. The innovation of proposed algorithm was that the coding of ROI mask was embedded into the SPIHT coding process. As the image and the mask information were coded synchronously, the bits stream could be truncated arbitrarily. Image quality evaluation method for ROI coding was also discussed. An image quality evaluation value called importance & area Weighted PSNR (WPSNR) was proposed by fully considering importance differences and area ratio between ROI and background. Experimental results show that proposed algorithm supports the compressing of multiple arbitrary shape ROI images from loss to lossless compression with configurable ROI priority. The generated bit stream could be truncated at any place, and the decoder can still decode image and ROI mask information well. Proposed algorithm was proved that it had a better performance than BbB-shift method. As the computational complexity of proposed coder was close to SPIHT, it can be used for ROI coding in low-rate applications or ROI preferential progressive transmission.