This paper presents a novel lossy compression scheme for medical images by using an incremental self-organized map (ISOM). Three neural networks for lossy compression scheme are comparatively examined: Kohonen map, multi-layer perceptron (MLP) and ISOM. In the compression process of the proposed method, the image is first decomposed into blocks of 8x8 pixels. Two-dimensional discrete cosine transform (2D-DCT) coefficients are computed for each block. The dimension of DCT coefficients vectors (codewords) is reduced by low-pass filtering. Huffman coding is applied to the indexes of codewords obtained by the ISOM. In the decompression process, inverse operations of each stage of the compression are performed in the opposite way. It is observed that the proposed method gives much better compression rates.