Lossless image compression continues to be the focus of the medical picture archiving system designers because of the possibility of reducing the bandwidth required to transmit medical images. The lossless Differential Pulse Code Modulation (DPCM) and Hierarchical Interpolation (HINT) have been suggested as solutions to this problem. However, there are limitations due to the inability of these schemes to adapt to local image statistics. Efforts to alleviate this problem can be seen in various adaptive schemes found in the literature. This paper introduces a new adaptive DPCM (ADPCM) scheme based on the shape of the region of support (ROS) of the predictor. The shape information of the local region is obtained through a universal Vector Quantization (VQ) scheme. The proposed lossless encoding scheme switches predictor type depending on the local shape. Simulation results show that improvements of about 0.4 bits/pixel over basic DPCM and 0.2 bits/pixel over HINT can be obtained. Comparison with lossless JPEG indicates that the proposed scheme can cope more easily with the changes in local image statistics. The computation required is moderate, since a universal VQ is used in encoding the shape information.