Comparison between two artificial intelligence models to discriminate cancerous cell nuclei based on confocal fluorescence imaging in hepatocellular carcinoma.

Background: Hepatocellular carcinoma (HCC) exhibits an exceptional intratumoral heterogeneity that might influence diagnosis and outcome. Advances in digital microscopy and artificial intelligence (AI) may improve the HCC identification of liver cancer cells.

Aim: Two AI algorithms were designed to perform computer-assisted discrimination of tumour from non-tumour nuclei in HCC.

Methods: Healthy livers and HCCs from commercially available tissue arrays were stained with an antibody against proliferating cell nuclear antigen and DRAQ5 dye with high affinity for double-stranded DNA, acquired by confocal microscopy imaging and then used to design machine learning (ML) and deep learning (DL) algorithms.

Results: Nuclei were segmented and then used to develop the Model 1 and Model 2 algorithms, using ML and DL respectively. Model 1 was trained with some texture nuclear features extracted using discrete wavelet transform and grey-level co-occurrence matrix. Model 2 was trained with the segmented images without any additional information. The comparative analysis of the models showed that DL was more effective than ML, achieving an average accuracy of 88 % in discriminating healthy from neoplastic nuclei in HCC samples.

Conclusion: Our research shows that AI techniques and nuclear fluorescent staining could be useful tools for automatically detecting HCC cells in liver tissues.