MAPPING FUNCTIONAL CHANGES IN THE EMBRYONIC HEART OF ATLANTIC SALMON POST VIRAL INFECTION USING AI TECHNIQUE

The problem of analysing videos of beating embryonic salmon hearts is addressed here with the aim of classifying infected and healthy embryonic salmon hearts. Determination of infection in a very early stage with simple means can reduce the cost of Salmon farming greatly and can boost the economy of countries dependent on Atlantic salmon fish farming. We converted data-dense videos into physically relevant single variable time-signal, namely area of heart as a function of time. This helped us derive simple physical attributes like spectrograms to describe the heart function in a more informative way. We used these spectrogram images with ensemble of deep learning algorithms to show successful classification of infected and healthy salmon hearts, with an average accuracy of 82%. We are unaware of such studies where simple but powerful physically relevant features are used for achieving this accuracy for videos with unconventional dynamics, such as beating heart.