Please use this identifier to cite or link to this item:
|Title:||AI for biological images||Authors:||Lim, Benjamin Kian Kuan||Keywords:||Engineering::Computer science and engineering||Issue Date:||2019||Abstract:||Advancement in technology within the last decade has led to the rapid development in the field of biological science. High-throughput of roughly 100,000 microscopic images can be yielded daily, through a motorized microscope available in the commercial market . The abundance of scientific data in the biological science field could help yield better analysis of lab experiments and tests. The aim of this project is to validate the use of deep learning methods to analyze cellular response of cancer cells in a controlled group. This is achieved by implementing a chosen deep learning methodology that is able to attain high accuracy in detection on a custom dataset of high resolution microscopic videos. In this experiment, the chosen deep learning methodology is the Faster Regional Convolutional Neural Network (R-CNN) using AlexNet and VGG-16 as the pre-trained model for comparison purpose. The network is trained on the custom dataset with 3600 training images and 867 testing images, with two object class, “Cancer cell” and “Cell rounding”. Results in this experiment shows that the Faster R-CNN method using VGG-16 as the pre-trained model achieve the highest mean Average Precision of 0.8516 after simple parameter tuning. In conclusion, even though this object detection using VGG-16 as the pre-trained model has achieved a high accuracy and performance result of 0.8516 mAP, it is still not accuracy enough to be used in a life and death critical field such as medicine. However, even with a high mean Average Precision score of 0.8516, it is still not accurate enough to be used in a life and death critical field such as medicine. Therefore, future works on the improvement on performance and accuracy is recommended.||URI:||http://hdl.handle.net/10356/78993||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Student Reports (FYP/IA/PA/PI)|
Page view(s) 1084
checked on Oct 25, 2020
checked on Oct 25, 2020
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.