Renal Cancer Cell Nuclei Detection from Cytological Images Using Convolutional Neural Network for Estimating Proliferation Rate

The Cytological images play an essential role in monitoring the progress of cancer cell mutation. The proliferation rate of the cancer cell is the prerequisite for cancer treatment. It is hard to accurately identify the nucleus of the abnormal cell in a faster way as well as find the correct proliferation rate since it requires an in-depth manual examination, observation and cell counting, which are very tedious and time-consuming. The proposed method starts with segmentation to separate the background and object regions with K-means clustering. The small candidate regions, which contain cell region is detected based on the value of support vector machine automatically. The sets of cell regions are marked with selective search according to the local distance between the nucleus and cell boundary, whether they are overlapping or non-overlapping cell regions. After that, the selective segmented cell features are taken to learn the normal and abnormal cell nuclei separately from the regional convolutional neural network. Finally, the proliferation rate in the invasive cancer area is calculated based on the number of abnormal cells. A set of renal cancer cell cytological images is taken from the National Cancer Institute, USA and this data set is available for the research work. Quantitative evaluation of this method is performed by comparing its accuracy with the accuracy of the other state of the art cancer cell nuclei detection methods. Qualitative assessment is done based on human observation. The proposed method is able to detect renal cancer cell nuclei accurately and provide automatic proliferation rate