Neural network-based species identification in venom-interacted cases in India

Abstract

India is home to a number of venomous species. Every year in harvesting season, a large number of productive citizens are envenomed by such species. For efficient medical management of the victims, identification of the aggressor species as well as assessment of the envenomation degree is necessary. Species identification is generally based on the visual description by the victim or a witness and is therefore quite likely to be erroneous. Symptomatic identification remains the only available method. In a previous published work, the authors proposed a classification table for snake species based on manifested symptoms applicable in Indian subcontinent. The classification table serves the purpose to a great deal but as a manual method it demands human expertise. The current paper presents a neural network-based symptomatic species identification system. A symptom vector is fed as input to the neural network and the system yields the most probable species as well as the envenomation severity as the output. The severity status can be very helpful in calculating the antivenom dosage and in deciding the species-specific prognostic measures for efficient medical management.