No abstract

Non-invasive bio-sensing is indispensable for safer patient care nowadays. In this regard, progressive developments for non-invasive haemoglobin (Hb) sensing used for anaemia diagnosis are based on digital photography or spectrometry, or spectrophotometric techniques. However, the analysis of these ailments by the non-invasive solutions stands challenging due to limiting satisfaction rate in various health conditions. An optimistic Near-Infrared (NIR) based spectrophotometric technique with an effective ML algorithm considering overcomplete influencing factors to overcome such challenges is presented in this manuscript. For this purpose, the data of 121 volunteers (19.27–55.46 years) has been employed to train and test the model using 5-Fold cross-validation with broad reference Hb values (8.2–17.4 g/dL). The highest accuracy is achieved using the mutual info regression feature selection technique with Support Vector Regression (SVR) and 3 input variables. Using this specific combination, cross-validation scores are obtained as; correlation coefficient (r_CV) = 0.796, standard deviation (SD_CV) = 1.069 g/dL, bias (Bias_CV)=-0.128 g/dL and limits of agreements (LoA_CV)= -2.223 to 1.967 g/dL. Moreover, variability between two standard devices is also presented to set the limits of agreement for the proposed technique. The mean scores to get the variability between two standard devices are observed as; r_mean = 0.970, SD_mean = 0.501 g/dL, Bias_mean = 0.209 g/dL and LoA_mean= -0.773 to 1.191 g/dL. Considerable precision in the range of ±1 g/dL is obtained while presenting the linear relationship between two standard devices. Therefore, the proposed technique is insurpassable and can expedite conventions for point-of-care (POC) applications in low-resource settings as well as in surgical sections which demand continuous monitoring.

No abstract

No abstract