Real-Time Image Error Detection in Knife-Edge Scanning Microscope

Research about the microstructure of the brain provides important information to help understand the functions of the brain. In order to investigate large volume, high-resolution data of mouse brains, researchers from Brain Network Lab (BNL) at Texas A&M University (TAMU) have been developing the Knife-Edge Scanning Microscope (KESM) in the past decade. The KESM can simultaneously section and image brain tissues at sub-micrometer resolution. However, malfunctions of the system can cause imaging errors, which make images fail to provide valid information. Moreover, malfunctions, especially due to obstructions (such as tissue fragments) in the light path of the system, result in continued cutting while the obstructions are present. Since KESM is generally not attended by a full-time human operator, this results in data loss.

To solve the problem, I developed an image error detection method to automatically find imaging errors in real-time. The method can detect errors by analyzing newly acquired images, report results to human operators and even stop the KESM cutting process if necessary so that data loss is avoided. The basic idea of the method is to solve error detection problem through image change detection algorithm as the images acquired by KESM are well-registered and they do not change too much from one slice to the next when there is no error. As a result, the method can detect imaging errors with 86% accuracy (F1-score) and finish a detection routine within 2 seconds, which is sufficient to achieve real-time detection. By integrating the error detection program into the KESM control system, the method enhanced the robustness of the system and reduced data loss.