This work introduces a novel methodological approach for the analysis of the Rey-Osterrieth Complex Figure (ROCF), a task widely employed in neuropsychological research, by modeling the drawing process as a time series. The aim is to obtain a compact spatio-temporal representation that addresses the limitations of traditional static-image approaches. The process involves two stages: segmentation of the drawing into strokes to capture spatial and temporal information, and their encoding into feature vectors using deep learning models such as BiLSTM-Autoencoder and ST2Vec. This encoding enables dimensionality reduction and standardization of the spatio-temporal representation of strokes. Validation is currently based on reconstruction accuracy, showing that both architectures preserve essential spatial and temporal aspects: BiLSTM-Autoencoder achieves lower RMSE, while ST2Vec better retains spatial structure. These results should be regarded as a foundational step towards the integration of encoded strokes into graph structures (GNN), enabling future analyses of handwritten drawings from a spatio-temporal perspective. While the long-term vision of this project includes applications in cognitive assessment, the present contribution is methodological in nature and establishes the basis for future extensions to clinically validated populations.