Abstract:
Machine Learning (ML) is a rapidly growing subfield of Artificial Intelligence (AI). It
is used for different purposes in our daily life such as face recognition, speech
recognition, text translation in different languages, weather prediction and business
prediction. In parallel, ML also plays an important role in medical domain such as in
medical imaging. ML has various algorithms that need to be trained with large volumes
of data to produce a well-trained model for prediction. Training of such algorithms with
small datasets is a challenging task. Data Augmentation (DA) refers to different
approaches that are used to increase the size of datasets. In this study, eight DA
approaches were used on medical images including; image rotation at 180o, rotation at
90o, noise, crop & scale, vertical flip, horizontal flip, shear and Gaussian blur. We used
a publicly available low-grade glioma tumor dataset obtained from Tumor Cancer
Imaging Archive (TCIA) repository. The dataset included 1961 MRI brain scan images
of low-grade glioma patients. Each DA approach was applied on original images in the
dataset to create new datasets by merging the results with the original dataset. You Only
Look Once (YOLO) version 3 model was trained on original dataset and the augmented
datasets separately. Supervisely with Tesla K80 GPU was used for YOLO v3 model
training on all datasets. The results showed that the DA techniques rotate at 180o and
rotate at 90o increased the accuracy of the results as 96% and 92% respectively. Crop &
Scale achieved 83% accuracy. Rest of techniques achieved <80% accuracy. Rotate at
180o and rotate at 90o are found as the best DA approaches for augmenting low-grade
glioma tumor medical imaging datasets.
