<p>Sweat sensing technology may enable continuous non-invasive patient monitoring. However, the clinical interpretation of biomarker concentrations in sweat may be hampered by their dependence on the sweat rate per gland. In this paper, we explore deep learning as a method to derive the sweat rate per gland, and compare its performance with an improved version of a rule-based approach, previously developed for the same purpose. To this end, we developed simulations mimicking the signal obtained through a digital sweat-sensing device, assumed to be (nearly) ideal, and using available knowledge on the pulsatile behavior of sweat glands. Additionally, we investigate the effect of noise and the employed loss function on the estimation of the number of active sweat glands (accuracy) and the sweat rate per gland (Mean Absolute Percentage Error (MAPE)). The obtained results show a MAPE of the sweat rate ranging from 3.1% to 35.6% and an accuracy in the estimation of the number of active glands ranging from 60.5% to 99.3%, depending on the different simulated conditions. Notably, the proposed deep learning approach achieved a good estimate of the sweat rate per gland throughout the physiological sweat rate range, outperforming the reference method, also in terms of computation time. This result demonstrates that the estimation of the sweat rate per gland through a digital sweat sensing device is feasible, facilitating the clinical interpretation of biomarker concentration in sweat and thus paving the way for sweat-based patient monitoring.</p>