Forewords

Oxynet is a scientific endeavor to improve automatic interpretation of cardiopulmonary exercise test data. To this, we are creating a set of algorithms that perform fast ventilatory threshold detection and synthetic data generation. The accuracy of these algorithms is compatible with that of clinical exercise physiologists.

Dataset statistics

Training dataset

In the following table, the anthropocentric characteristics of the individuals included in the training dataset are presented.

Train dataset: anthropometric characteristics
Selection based on fitness level
Fitness group	Age (y)	Age SD	Height (m)	Height SD	Weight (kg)	Weight SD
Low	55.66	12.03	1.66	0.09	108.71	22.52
Medium	53.66	9.74	1.69	0.08	71.13	10.35
High	25.95	8.73	1.74	0.08	66.30	8.88

In the following table, the aerobic fitness characteristics of the individuals included in the train dataset are presented.

Aerobic fitness characteristics
Selection based on fitness level
Fitness group	VO2PEAK (mlO2/min)	VO2PEAK SD	VO2PEAK-N (mlO2/min/kg)	VO2PEAK-N SD
Low	1,923.68	513.28	18.02	4.46
Medium	2,118.09	766.49	29.71	9.29
High	4,357.65	753.66	65.84	7.98

Test dataset

In the following table, the anthropometric characteristics of the individuals included in the test dataset are presented.

Train dataset: anthropometric characteristics
Selection based on fitness level
Fitness group	Age (y)	Age SD	Height (m)	Height SD	Weight (kg)	Weight SD
Low	55.68	12.03	1.66	0.09	108.73	22.51
Medium	53.66	9.74	1.69	0.08	71.13	10.35
High	25.95	8.73	1.74	0.08	66.30	8.88

In the following table, the aerobic fitness characteristics of the individuals included in the test dataset are presented.

Aerobic fitness characteristics
Selection based on fitness level
Fitness group	VO2PEAK (mlO2/min)	VO2PEAK SD	VO2PEAK-N (mlO2/min/kg)	VO2PEAK-N SD
Low	1,922.86	513.03	18.00	4.46
Medium	2,118.09	766.49	29.71	9.29
High	4,357.65	753.66	65.84	7.98

Body mass index for different fitness levels

The following box plot, gives an idea about how data are distributed around the average values of body mass index (BMI). Classification has been made by gender and aerobic fitness level.

BMI and fitness group.

Peak oxygen uptake for different fitness levels

The following box plot, gives an idea about how data are distributed around the average values of peak oxygen uptake during the cardiopulmonary test. Classification has been made by gender and aerobic fitness level.

Peak-VO2 and fitness group.

Oxynet performance

In this section, the performance indices of the ventilatory detection deep learning algorithm are provided.

Summary of the results in tables

In this table, differences are highlighted by means of the root mean square (RMS) error. Values are given in seconds of difference between Oxynet estimations and experts’ estimations. Whereas it is not specified, statistics are reported separately for individuals falling in different “aerobic fitness” groups:

Low fitness level
Medium fitness level
High fitness level

RMS error Oxynet VS Experts
Selection based on fitness level. Time-based thresholds.
Fitness group	RMS VT1 (s)	RMS VT1 SD	RMS VT2 (s)	RMS VT2 SD
Low	10.79	8.73	11.70	14.53
Medium	28.80	16.25	12.70	7.88
High	14.75	12.89	8.66	7.97

In this table, differences are highlighted by means of the RMS as above, but this time differences are provided in mlO2/min between Oxynet estimations and experts’ estimations.

RMS error Oxynet VS Experts
Selection based on fitness level. VO2-based thresholds.
Fitness group	RMS VO2VT1 (mlO2/min)	RMS VO2VT1 SD	RMS VO2VT2 (mlO2/min)	RMS VO2VT2 SD
Low	40.67	48.11	28.68	30.97
Medium	60.00	79.83	31.30	37.62
High	104.24	133.36	52.59	82.86

Graphical representations

Box plots are used in this context to highlight data distribution, central tendency and dispersion.

First ventilatory threshold: time estimation for expert and Oxynet.

Second ventilatory threshold: time estimation for expert and Oxynet.

Bland-Altman plot: first ventilatory threshold (Experts VS Oxynet).

Bland-Altman plot: second ventilatory threshold (Experts VS Oxynet).

RMS first ventilatory threshold (Experts VS Oxynet).

RMS second ventilatory threshold (Experts VS Oxynet).