Which signal can I use?
The literature review or some expertise will give you the reference method to measure (blood samples for example) or score the status (infection or not) of a target variable.
Before being able to efficiently use sound as an indicator of health status, it is necessary to detect which sound represents an infection or not, at which level it could be considered as infected or not, and if possible, the level of infection.
Two algorithms have therefore to be developed
From your record, you need to sort out what is a cough from what is a sneeze, in order to be able to precisely identify them.
While labelling your initial data, you try to find that. During this process, a detailed audio-visual analyses is performed, in order to understand all variations in the measured field data to develop an algorithm calculating the feature variable.
In present study using cough in pigs, a first algorithm development is necessary to transform the crude measurement (sound) into valuable information, i.e. the number of coughs.
Having the number of coughs is not enough, since you need to have the information of what is the normal or abnormal situation, i.e. the number of coughs. A second algorithm is then necessary followed by the final validation.
Validation is a crucial step, and will confirm or not the interest of your method, based on a comparison with the “gold reference”.
In all cases, you will need to calculate sensitivity, specificity and accuracy of your method, defined as follows :
Sensitivity = (number of True Positives, TP) / (number of TP + number of False Negatives, FN)
Specificity = (number of True Negatives, TN) / (number of TN + number of False Positives, FP)
Accuracy: (TP + TN)/(TP+FP+FN+TN).
For example: cough monitoring
