Abstract

Peripheral Blood Smear (PBS) analysis is a vital routine test carried out by hematologists
to assess some aspects of humans’ health status. PBS analysis is prone to human errors and
utilizing computer-based analysis can greatly enhance this process in terms of accuracy
and cost. Recent approaches in learning algorithms, such as deep learning, are data hungry,
but due to the scarcity of labeled medical images, researchers had to find viable alternative
solutions to increase the size of available datasets. Synthetic datasets provide a promising
solution to data scarcity, however, the complexity of blood smears’ natural structure adds
an extra layer of challenge to its synthesizing process. In this thesis, we propose a method-
ology that utilizes Locality Sensitive Hashing (LSH) to create a novel balanced dataset of
synthetic blood smears. This dataset, which was automatically annotated during the gener-
ation phase, covers 17 essential categories of blood cells. The dataset also got the approval
of 5 experienced hematologists to meet the general standards of making thin blood smears.
Moreover, a platelet classifier and a WBC classifier were trained on the synthetic dataset.
For classifying platelets, a hybrid approach of deep learning and image processing tech-
niques is proposed. This approach improved the platelet classification accuracy and macro-
average precision from 82.6% to 98.6% and 76.6% to 97.6% respectively. Moreover, for
white blood cell classification, a novel scheme for training deep networks is proposed,
namely, Enhanced Incremental Training, that automatically recognises and handles classes
that confuse and negatively affect neural network predictions. To handle the confusable
classes, we also propose a procedure called "training revert". Application of the proposed
method has improved the classification accuracy and macro-average precision from 61.5%
to 95% and 76.6% to 94.27% respectively.
In addition, the feasibility of using animal reticulocyte cells as a viable solution to com-
pensate for the deficiency of human data is investigated. The integration of animal cells is implemented by employing multiple deep classifiers that utilize transfer learning in differ-
ent experimental setups in a procedure that mimics the protocol followed in experimental
medical labs. Moreover, three measures are defined, namely, the pretraining boost, the
dataset similarity boost, and the dataset size boost measures to compare the effectiveness
of the utilized experimental setups. All the experiments of this work were conducted on
a novel public human reticulocyte dataset and the best performing model achieved 98.9%,
98.9%, 98.6% average accuracy, average macro precision, and average macro F-score re-
spectively.
Finally, this work provides a comprehensive framework for analysing two main blood
smears that are still being conducted manually in labs. To automate the analysis process,
a novel method for constructing synthetic whole-slide blood smear datasets is proposed.
Moreover, to conduct the blood cell classification, which includes eighteen blood cell types
and abnormalities, two novel techniques are proposed, namely: enhanced incremental train-
ing and animal to human cells transfer learning. The outcomes of this work were published
in six reputable international conferences and journals such as the computers in biology and medicine and IEEE access journals.