Automatic quantitative and morphometric analysis of muscle fibers.

In this work, we propose an automated solution to support researchers of histological sciences labs on the analysis of muscle fibers microscopic images. For this purpose, we developed a dataset carefully curated composed of semi-seriated muscle samples. The manual annotation of these images is laborious and tiring, and the results may become compromised as the human gets tired. Thus, we propose a framework to address this problem which starts with a preprocessing step, aiming to mitigate discrepancies regarding the color variation frequently found in this type of image, followed by the use of Mask R-CNN, to automatically perform muscle fibers segmentation and also to extract the smallest diameter of these fibers, which is an important measure for further analysis. Experiments showed that there is no statistically significant difference between the results obtained manually and those obtained using our proposal. The dataset used in the experiments is also a contribution of this work.