Automated Nuclei Segmentation Using Shearlet Based Unsharp Masking

Accurate nuclei segmentation is essential in pathology since evaluations and diagnoses are mostly dependent on the measurement, identification, and counting of nuclei. Automated segmentation in digital pathology is a difficult task due to several factors, including, hazy nucleus boundaries, low contrast between nuclei and background, varying nuclei sizes and shapes, and imaging modalities. In this work, a new nuclei segmentation method is proposed which incorporates the shearlet-based unsharp masking (SBUM) technique to pre-process the source images together with the CNN-based U-Net architecture. The SBUM method primarily enhances the source image's high-frequency details. The effectiveness of a developed nuclei segmentation technique is compared to the existing techniques using a common dataset. The developed method achieves IOU, Accuracy, F1Score, and Precision values as 0.8146, 95.93%, 0.8953, and 0.8725 respectively. The empirical findings demonstrate that the developed method outperforms the existing methods.