Abstract
[Introduction] A tail half of a dissected Planarian (Dugesia Dorotocephala) can show a previously-conditioned response after its regeneration, possibly even before maturation of the head and the neural ganglia (Shimojo, et al., VSS ’21). Our results, however, were mainly based on eye examination of their responses to an Electric Shock (ES)(immediate contraction) and UV light (wiggling) before/after conditioning. We now report more objective classification based on machine learning (DNN) to confirm it. [Method] After we conditioned intact Planarians in a Pavlovian procedure with an ES as the unconditioned stimulus (UCS), and weak UV light as the conditioned stimulus (CS), we dissected their bodies into halves. Starting from the 2nd day after dissection, we presented the CS 3 times daily while video-recording the responses. The procedures were automated by an Arduino microcontroller. The recorded responses were then classified by DNN. We used the DNN model VGG16 pretrained by ImageNet for extracting features from images and the classifier. Its three final layers were replaced and trained with transfer learning, fed with labeled video frames extracted from each second of each video. We trained the network with 211 responses to ES and 118 to UV light before conditioning/dissection. The cross-validated accuracy was 83.6%. We then let it classify responses of the tail halves to the UV light. [Results and Discussion] 99 recorded responses to UV from 20 individual conditioned tails across day 2-3 were analyzed. 96.8 % were classified as “ES-induced” (against 22.0% wrongly classified as “ES-induced” for unconditioned samples under UV), therefore qualified to have shown the “Conditioned Response” (p<3.06E-30). Thus, the classification based on DNN provided objective evidence that the Planarian can conserve and reveal a learned response without the head/ganglia. The results were consistent with the distributed nature of neural pathways and UV sensors (Shettigar, et al., Sci.Adv., 2017).