An Artificial Intelligence-Based Image Processing Framework for Gene Expression Analysis and Disease Prediction

Abstract

Artificial intelligence has increasingly been an essential aid to combining biomedical image analysis with molecular-level data to detect disease correctly. Nevertheless, the uncertainty in image quality, segmentation, inconsistency in features, and heterogeneity in gene-expression may diminish the accuracy of the traditional prediction models. This paper presents an artificial intelligence-based image-processing system to analyze gene expression and predict diseases. The model combines image preprocessing, segmentation via deep-learning, feature-extraction, estimation of gene-expressions, robust optimization based on minimax, and classification of diseases in a single computational framework. Noise reduction, contrast enhancement and intensity normalization are used to improve biomedical images and then disease-relevant regions are segmented and informative morphological, texture, intensity and deep features are obtained. These characteristics are then utilized to estimate pattern of gene-expression and help in classification of disease. A minimax optimisation technique is added to minimize worst-case prediction loss with respect to both image and feature perturbation to enhance model robustness. Image quality- PSNR and SSIM, segmentation- Dice score and IoU, gene-expression estimation- correlation coefficient, RMSE and MAE, disease prediction- accuracy, precision, sensitivity, specificity, F1-score and ROC-AUC are used in the evaluation of performance. The suggested framework can offer a well-integrated and powerful computational way to connect the biomedical image features to gene expression data and computer-aided diagnosis of illnesses.