Turkish Journal of Electrical Engineering and Computer Sciences




Early diagnosis of colorectal cancer lengthens human life and is helpful in efforts to cure the illness. Histopathological inspection is a routinely utilized technique to diagnose it. Visual assessment of histopathological images takes more investigation time, and the decision is based on the individual perceptions of clinicians. The existing methods for colorectal cancer classification use only spatial information. However, studies on the spectral domains of information are lacking in the literature. Therefore, the performance of the existing techniques is moderate. To improve the performance of colorectal cancer classification, this work proposes a unique hybrid domain hand-crafted feature formulated using scale-invariant feature transform and mel-cepstrum domain features. The developed hand-crafted features use spatial as well as spectral information. Furthermore, the developed hand-crafted features are given as input to a newly developed 1D multiheaded convolutional neural network (1D MHCNN) for the classification of colorectal tissue utilizing histopathological images. The performance of the proposed network is compared with other existing methods. Based on the experiments, the proposed network performed with accuracy of 96.80%, specificity of 99.76%, precision of 97.12%, sensitivity of 96.64%, F1 score of 0.9688, and area under the curve of 0.9820. The proposed approach may be utilized to improve clinical diagnosis measurement performance.


Histopathological images, colorectal cancer, mel-cepstrum, multiheaded convolutional neural network, hand-crafted feature

First Page


Last Page