Imaging Processing and Machine Learning in Clinical Microscopy
In many clinical laboratory and pathology testing procedures, visual examination of microscopic slides is needed, e.g., to classify disease developments, to detect the interactions of micro-organisms, to assess the effectiveness of drugs, to determine cell viability and proliferation, etc. These tasks can be found in many important clinical applications, including cancer research, hematology, pharmacology, and genetic testing.
Traditionally, these tasks are performed manually by a qualified laboratory technician. In many cases, the tasks effectively require direct visual inspection, or some form of calibrated estimation, e.g., using a hemocytometer under microscope. Evidently, this approach is not only time-consuming and tedious, but also prone to errors due to human subjectivity and fatigue. Moreover, this ineffective approach is unsuitable to accommodate a large number of specimen samples. In fact, many cell-based research studies require rapid turn-around time, with high accuracy and reliability, in order to provide useful interpretation and analysis for clinical effectiveness.
Recent promising solutions towards clinical microscopy are based on automated image processing and analysis. These methods provide analyses and interpretation of images obtained from a digital microscope, equipped with a camera. For example, suitable image processing and segmentation are performed to identify and locate various biological cells. Using advanced machine learning, it is also possible to classify various cell types effectively, e.g., viable vs. dead cells, cells with vs. without antibodies. Growth factors and drug treatments also result in visual feature changes, and as such can be characterized automatically based on the collected images. At the same time, there are many challenges that need to be addressed to make this approach a reliable and economically feasible solution. For instance, as in many image-based problems, the presence of imaging artifacts, focusing errors, and source contamination, can create severe difficulties in the image processing procedure. Therefore, suitable pre-processing methods will need to be designed and tested under practical laboratory settings, with the advice of qualified experts in biochemistry and clinical pathology, particularly from the Saskatchewan Cancer Agency. The synergy of engineering analysis and health science expertise will be crucial in delivering an effective image-based clinical microscopy framework.
In summary, this project addresses an important need in clinical laboratory science and pathology, providing an alternative framework for conducting clinical microscopy, that is not only more rapid compared to manual methods, but also more reliable and accurate. In addition, by allowing for digitization and computer-based analyses, many additional tasks can also be envisioned. For instance, pattern recognition and machine learning can be performed to deliver relevant visual representation, and detect long-term trends based on a collection of historical microscopic images. These tasks would not be possible with manual microscopic methods, especially with the advent of the big data era. The success of this project would have important impact for clinical studies, which in turn will lead to improved disease characterization and treatment solutions. These are all important accomplishments in improving the quality of human life, by offering health researchers a crucial tool in the fight against deadly diseases such as cancer.
