Page 17 - Read Online
P. 17
Page 101 Liu et al. Art Int Surg 2024;4:92-108 https://dx.doi.org/10.20517/ais.2024.19
Table 3. Performance of our multi-class classification model observed when we ablate the inclusion of key individual joints that are
central to modeling lower-arm orientations and computing visual attention
Pelvic joints Arm joints Cranial joints
All Wrists Elbows Eyes Head Ears Recall↑ Precision↑ F1↑ AUPRC↑
√ √ - - - - 0.77 0.74 0.75 0.78
√ √ √ - - - 0.73 0.72 0.72 0.78
√ √ √ √ - - 0.80 0.79 0.79 0.80
√ √ √ √ √ - 0.79 0.79 0.79 0.82
√ √ √ √ √ √ 0.83 0.82 0.81 0.85
- √ √ √ √ √ 0.77 0.75 0.76 0.82
Embeddings are modeled as 3D joint positions. Bolding indicates a top score. AUPRC: The area under the precision-recall curve.
Figure 3. Comparisons of positional heat maps among tracklets engaging in walking movements (A-C) and tracklets engaging in hand-
tool interactions (D-F). Tracklets engaging in walking movements (A-C) are more positionally dispersed, represented by the wide
spread of their positional heat signature, while tracklets engaging in hand-tool interactions (D-F) are more visibly concentrated in a
position close to the operating table.
