Convolutional neural networks have shown to achieve superior performance on\nimage segmentation tasks. However, convolutional neural networks, operating as\nblack-box systems, generally do not provide a reliable measure about the\nconfidence of their decisions. This leads to various problems in industrial\nsettings, amongst others, inadequate levels of trust from users in the model's\noutputs as well as a non-compliance with current policy guidelines (e.g., EU AI\nStrategy). To address these issues, we use uncertainty measures based on\nMonte-Carlo dropout in the context of a human-in-the-loop system to increase\nthe system's transparency and performance. In particular, we demonstrate the\nbenefits described above on a real-world multi-class image segmentation task of\nwear analysis in the machining industry. Following previous work, we show that\nthe quality of a prediction correlates with the model's uncertainty.\nAdditionally, we demonstrate that a multiple linear regression using the\nmodel's uncertainties as independent variables significantly explains the\nquality of a prediction (\\(R^2=0.718\\)). Within the uncertainty-based\nhuman-in-the-loop system, the multiple regression aims at identifying failed\npredictions on an image-level. The system utilizes a human expert to label\nthese failed predictions manually. A simulation study demonstrates that the\nuncertainty-based human-in-the-loop system increases performance for different\nlevels of human involvement in comparison to a random-based human-in-the-loop\nsystem. To ensure generalizability, we show that the presented approach\nachieves similar results on the publicly available Cityscapes dataset.\n