 Cerebellar ataxia, CA, is a disorder characterized by impaired coordination of movements due to damage or malfunctioning of the cerebellum. This condition can affect eye, speech, trunk, and limb movements, and conventional machine learning approaches have been used to objectively diagnose and quantify the severity of CA. However, these approaches require large data sets and raise privacy concerns. To address these issues, we propose an image transformation-based approach using federated learning to diagnose CA. Our method uses motion capture sensor data collected from four different clinics, which is transformed into three visual features, recurrence plot, mel-spectrogram, and Poirier plot, and then fed into a MobileNet V2 model. Experiments show that our approach achieves higher accuracy than other methods, while also protecting patient privacy. This article was authored by Tang No, Dinh, Singh, Nguyen, Pobudu, and Patherana, and others. We are article.tv. Links in the description below.