Objective: The aim of this study was to describe the stages of learning Kohonen's self-organizing maps applied to scintigraphy imaging in order to perform classification for medical diagnostic aid. Method: To achieve these goals, the neurons, arranged on a regular grid, are connected to each other by a neighbor relationship, which creates the topology of the map. The input layer consisted of pixels from the scintigraphy images. Results: During the iteration rounds of learning, we have seen a deployment of neurons on the nodes of the map that becomes more and more important. And it is the same for the winning neurons. After 750 iterations, the Davies Bouldin index attests to the end of the training with a quantization error that goes from 0.175 at the beginning of the training to 0.0225 at the end of the training. After this study, we find that neurons 41, 62, 121, 101 and 145 have captured most of the data with a peak uptake achieved by neuron 41 which has captured 1048 data. This individualizes the class of high intensities undoubtedly corresponding to metastatic hyperfixations. Conclusion: This innovative method could undoubtedly be integrated into the link in the chain highlighting periarticular metastases in developing countries, most of which do not have a SPECT-CT.