An arbitrary shape ROI coding method based on modified Set Partitioning in Hierarchical trees (SPIHTs) is proposed to solve the problems of arbitrary shape Region of Interest(ROI) coding algorithms in low reconstructed image quality and inefficiency at a low bit rate. The ROI mask is described by combining the similarity of spatial position among wavelet coefficients with the specific wavelet filter at an extremely low bit rate, which lays a foundation for improving coding efficiency. The modified SPIHT partitioning structure is used to improve the coding efficiency of sorting pass by grouping more ROI entries together. A quantization method based on each wavelet subband optimizes the output bit stream in ROI encoding and improves the reconstructed image quality by choosing the proper threshold for each subband. Experiment results show that the proposed method can support multiple arbitrary shape ROIs and can describe the ROI mask information of the whole image at very low bit rate (less than 0.04 bit/pixel). Furthermore, at the bit rate less than 0.5 bit/pixel,the proposed algorithm has improved its Peak Signal to Noise Ratio(PSNR) by 2-7dB,and reduced the encoding time above 30% as compared with those of Multiple Subband Bitplane Shift (MSBShift) algorithm based on JPEG2000. It is concluded that the proposed method has higher reconstructed image quality, faster coding speed and is fit for image application at the low bit rate.