Test set. Confusion matrix. Tables compare data provided by the veterinarians (“ground-truth annotation”) with those provided by the baseline DL method (A) and by the ad hoc trained DL method (B). The diagonal elements represent the correct prediction; the darker the blue color, the more correct the prediction of the DL-based method. The ad hoc trained method (B) correctly predicted a very high number of healthy (48/50), class IV (46/50) and class III (42/50) half carcasses. The prediction was less effective for class II lesions (35/50). Overall, the ad hoc trained DL-based method provided better results as compared with those given by the baseline DL method.