Supervised Learning-based Nuclei Segmentation on Cytology Pleural Effusion Images with Artificial Neural Network
Abstract: Automated segmentation of cell nuclei is the crucial step towards computer-aided diagnosis system because the morphological features of the cell nuclei are highly associated with the cell abnormality and disease. This paper contributes four main stages required for automatic segmentation of the cell nuclei on cytology pleural effusion images. Initially, the image is preprocessed to enhance the image quality by applying contrast limited adaptive histogram equalization (CLAHE). The segmentation process is relied on a supervised Artificial Neural network (ANN) based pixel classification. Then, the boundaries of the extracted cell nuclei regions are refined by utilizing the morphological operation. Finally, the overlapped or touched nuclei are identified and split by using the marker-controlled watershed method. The proposed method is evaluated with the local dataset containing 35 cytology pleural effusion images. It achieves the performance of 0.95%, 0.86 %, 0.90% and 92% in precision, recall, F-measure and Dice Similarity Coefficient respectively. The average computational time for the entire algorithm took 15 mins per image. To our knowledge, this is the first attempt that utilizes ANN as the segmentation on cytology pleural effusion images.