Enhanced Histopathological Image Reconstruction and Classification Using Multi-Input Super-Resolution and Lightweight Neural Networks

Liver cancer is one of the leading causes of mortality in cancer-related diagnoses in previous years. The mortality rate can be reduced if the cancer is identified at an early stage. In the early stages, the images are acquired through radiography imaging. However, in critical cases, histopathological imaging is used. In these cases, extreme care is to be taken to avoid any misclassification. The histopathological images are high-resolution images; however, in cases where image quality is lost, classification accuracy will be degraded. In this paper, a multi-input super-resolution neural network (MISRNN) is proposed to restore high-resolution images from low-resolution images. To carry out the proposed work, the histopathological images of four classes were collected from a private hospital. To simulate the real-world scenario, the low-resolution images of factors ×2, ×4, and ×6 are obtained through the bicubic interpolation technique. To evaluate the performance of the proposed model, MISRNN, the quantitative metrics PSNR and SSIM are used. The proposed MISRNN achieved the PSNR values of 39.12, 33.98, and 31.02 dB and SSIM values of 0.948, 0.868, and 0.807 on the ×2, ×4, and ×6 images, respectively. The reconstructed super-resolution images are used for classification. The performance of the proposed classification model is improved by the reconstructed super-resolution images. The proposed model can classify the reconstructed super-resolution histopathological images with an accuracy value of 96.7%. The proposed methodology, super-resolution followed by classification, can be used as a benchmark for further research.