Bones, Algorithms, and Behaviour : AI-based 2D:4D Measurements and Links to the Experience of Flow and to Physical Activity

Sammanfattning

This study aimed to fill a methodological gap in 2D:4D ratio research by developing an artificial intelligence (AI) model that recognizes endpoints (EPs) of bones in radiographic images, and thus measures finger lengths and determines 2D:4D ratio. Additionally, this study looked at connections between 2D:4D ratio, experience of flow, and physical everyday activity (PEA). Participants filled out questionnaires on flow and PEA and their hands were radiographed and measured. A deep learning (DL) based model was trained on 220 manually labelled radiographic images and is based on two parts, as the first one focussed on segmenting the four selected fingers and the second one on predicting EPs in these fingers. The architectures are based on a U-Net structure, as the segmentation model is also based on a residual network (ResNet) structure. The AI model demonstrated a poor to fair performance in identifying EPs, when predictions on 26 unlabelled radiographic images were compared to the manually labelled data. This is most likely due to the small training data set. Resorting to manually labelled data, no negative associations were found between 2D:4D ratio, flow and PEA. Finally, further research is needed to perfect an AI model, and other aspects of flow and PEA could be considered further.