Autonomous interpreting peripheral blood film based on deep learning algorithm

• Main Author: Salehuddin, Nur Anisah
• Format: Thesis
• Language: English
• Published: 2020
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utm.my/id/eprint/93018/1/NurAnisahSalehuddinMSKE2020.pdf; http://eprints.utm.my/id/eprint/93018/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:135903
• Institution: Universiti Teknologi Malaysia

The peripheral blood film (PBF) is a laboratory work-up that involves cytology of peripheral blood cells smeared on a slide. As basic as it is, PBF is invaluable in the characterization of various clinical diseases as the PBF is an informative haematological tool at the clinician’s disposal in screening, diagnosis and monitoring of disease progression and therapeutic response. Common clinical indication for PBF includes unexplained cytopenia, anaemia, unexplained jaundice, chronic myeloid leukaemia, suspected organ failure such as renal disease, liver failure, lymphoma and chronic lymphocytic leukaemia. PBF can only be interpreted under the microscope. A quick assessment of a PBF can be made within 3 minutes by a skilled laboratory physician but an abnormal film would require a longer time for wider view and differential cell counts. In addition, with the increasing amount of PBF screening (up to hundreds) samples requested per day, it is impossible for the laboratory physician to finish up the PBF screening within the given time frame. Besides, this conventional method tends to give inconsistent outcome as well as poor accuracy due to the significant level of inter-observer variation in grading. In Malaysia particularly, the PBF screening only available in selected General Hospital who has Hematopathology unit. Thus, all PBF samples from Klinik Kesihatan and District Hospital will be sent out to this hospital. The process itself is time consuming and tedious. Therefore, this project is aimed for the PBF to be analysed by a system that could differentiate the component on PBF which are, red blood cell (RBC), white blood cell (WBC) and platelets quantitively. Faster R-CNN algorithm for object detection is implemented as the deep learning framework for training, validating and testing the PBF images. The framework is built by integrating the Keras object detection package on top of backbone, Tensorflow library with Python as the programming language.