Freezing of gait assessment with inertial measurement units and deep learning: effect of tasks, medication states, and stops

Table 8 Model comparison results in terms of training with/without trials containing stops

Model	Test data	Segment-F1@50		Sample-F1
Model	Test data	Mean	t-test (#pair)	Mean	t-test (#pair)	#FP-Stop
Model_Clinical	Non-stop	0.60	t(12) = 0.16, p = 0.874	0.67	t(12) = 0.02, p = 0.979	N.A.
Model_Stop	Non-stop	0.59	t(12) = 0.16, p = 0.874	0.67	t(12) = 0.02, p = 0.979	N.A.
Model_Clinical	Stopping	0.40	t(12) =− 5.31, p < 0.005	0.46	t(12) = − 4.39, p < 0.005	74/210
Model_Stop	Stopping	0.59	t(12) =− 5.31, p < 0.005	0.63	t(12) = − 4.39, p < 0.005	16/210

We investigated the effect of including or excluding stopping periods in FOG detection by comparing models trained with (i.e., Model_Stop) and without stopping trials (i.e., Model_Clinical). The third and fourth column depicts Segment-F1@50 averaged over all 12 subjects and the paired t-test result. The fifth and sixth columns depict the Sample-F1 averaged over all 12 subjects and the paired t-test result. The seventh column depicts the number of stops detected as FOG with respect to the total number of stops (#FP-Stop). N.A. was shown for #FP-Stop when testing on trials without stopping as it would not be possible to detect stopping as FOG

ISSN: 1743-0003