Artificial intelligence assisted drop pattern analysis and RNAseq profiling for early diagnosis and follow-up of bladder cancer

Bladder cancer is one of the most common cancer types in the urinary system. Current bladder cancer diagnosis and follow-up techniques are time-consuming, expensive, and invasive. The gold standard for the diagnosis of bladder cancer in clinical practice is invasive biopsy followed by histopathological analysis. In recent years, costly tests involving bladder cancer biomarkers were developed, but these tests have high false-positivity and false-negativity rates, limiting their reliability. Hence, there is an urgent need for the development of novel and practical diagnostic tests. In this thesis, by analyzing droplet patterns of blood and urine samples from patients, we developed a deep learning- and artificial intelligence-assisted quick, cheap, and reliable diagnosis method. Droplet pattern analysis of evaporated blood or urine deposits was performed using patient and normal control samples. Our proposed AI-assistant model (ResNet-18 pre-trained ImageNet) can be systematically applied across droplets, enabling comparisons to reveal shared spatial behaviors and underlying morphological patterns, which precisely differentiate patient-derived samples from controls with high accuracy. The innovative diagnostic method has been presented based on the recognition and classification of complex patterns formed by dried urine or blood drops under different conditions. Our results indicate that AI-based model have a great potential for a non-invasive and accurate diagnosis of bladder cancer. RNA sequencing was also performed to identify potential candidate markers for bladder cancer in different gene classes, including up and downregulated genes, gradient-increased or decreased genes, case-specific genes, and upregulated genes encoding secreted proteins. According to the results, various novel genes have been found to be candidate markers that can be used for bladder cancer diagnosis. For example, the OVOL2 gene was found to be a disease-free marker for bladder cancer. In addition, eleven genes encoding secreted proteins were found to be potential secreted candidates from bladder tumors. Also, the integration of blood or urine droplet patterns with these secreted proteins was performed to investigate the possible contribution of these secreted proteins to droplet patterns. It was found that expression levels of these genes were differentiated among patients' blood and urine droplet patterns may be an alternative perspective to determine patients who have bladder tumor variations. Two genes were also chosen from RNAseq outputs for molecular analysis. One of them was knocked out (KO) using CRISPR/Cas9 system. The KO-T24 bladder cell line has shown increased spheroid diameter in three-dimensional cell culture system compared to wild type. Another chosen gene was also evaluated for mRNA expression level in tumor samples, and the expression level of the gene was found to be increased in the tumors compared to control tissue. In conclusion, the innovative diagnostic method has been presented based on recognizing and classifying complex patterns formed by dried urine or blood drops under different conditions. Our results indicate that AI-based systems have great potential for a non-invasive and accurate diagnosis of bladder cancer. Determined candidate genes from RNA sequencing will be expected to understand the molecular biology of bladder cancer and discover new therapeutic perspectives in bladder cancer management.