Automatic multilevel image thresholding segmentation using hybrid bio-inspired algorithm and artificial neural network for histopathology images

Automated medical imagining is growing rapidly for advanced clinical treatment and intervention in medical diagnosis. The segmentation of nuclei in digital histopathology is considered the most crucial aspect in diagnosis and evaluating the severity of disease. Therefore, in this paper, an automated nuclei segmentation method has been introduced for the histopathological images. The proposed segmentation method uses a new hybrid algorithm of lion optimization and cat swarm optimization to provide an optimal threshold value for efficient multi-level image thresholding segmentation. Moreover, a new fitness function using Otsu's function and Yager's entropy has also been presented for robust results. Extensive simulations are examined to determine the performance of the proposed hybrid algorithm. Firstly, evaluation on CEC 2017 benchmark suite are performed using three quality metrics, namely average fitness value, peak signal-to-noise ratio, and structural similarity index. Secondly, for the performance analysis of developed segmentation method, two Histopathological data sets are used, namely breast Histopathological images and lung Histopathological images. Furthermore, to find the optimum number of clusters, artificial neural network has been used and its performance is measured in terms of the accuracy, sensitivity, and specificity. The classifier returned an accuracy of 71.4%, sensitivity of 93.23%, and specificity of 93.12% for breast Histopathological images. For lung Histopathological images, the accuracy, sensitivity, and specificity were 66.17%, 89.53%, and 90.62% respectively. Moreover, the proposed method is also validated using the ground truth images, acquired from expert pathologist, by calculating dice coefficient and Jaccard coefficient. The findings and observation show that the proposed algorithm provides promising and significant result.

Automated summary

This paper presents an automated nuclei segmentation method for histopathological images using a hybrid algorithm of lion optimization and cat swarm optimization. The proposed method achieves efficient multi-level image thresholding segmentation through the optimal threshold value provided by the hybrid algorithm. Extensive simulations and analysis on benchmark suites and histopathological datasets demonstrate the performance and effectiveness of the proposed method.