Detection of cell nucleus is an essential and fundamental task in a wide range of biomedical studies and clinical applications, paying a special attention to the disease diagnosis and monitoring. Automation of this process is still challenging.
First, target cells, in many cases being very small, are surrounded by clutter represented by complex histological structures like capillaries, adipocytes, collagen, etc. Second, the amount of target cells may range from tens to thousands in a typical high resolution microscopy image.
Furthermore, the problem becomes more complicated because of the diversity of nuclear morphology and different imaging technology applied. These challenges render the cell nucleus detection far from being solved. Thus, the project team aims to develop the model based on the artificial intelligence used to detect cell nucleus in fluorescence images.
Reearches focuse on images containing noise, touching and partially overlapping nuclei, and possible observed under different lighting conditions. Recent studies indicate that deep learning may yield superior accuracy in the field of digital pathology. Therefore, the technique of convolution neural networks (CNN) will be applied where the filters will be modelled in advance (prior to the learning of network) and then embedded in the architecture. This concept is motivated by a need to have this technology effective to use it in real-world laboratories.
Next, recently appeared deep learning technology named as capsule neural network will be investigated for this type of images in order to develop the architecture invariant to the cell nucleus form, staining technique used and illumination. The performance of model to be developed will be compared with already existing CNN architectures such as AlexNet, GoogleNet, VGGNet, U-Net, FCNN, etc.