Objectives: Urinary bladder cancer is one of the most common malignancies of the urinary tract. It begins when cells of the bladder mucosa start to grow uncontrollably with the tendention to spread. Such a spread can cause cancer to expand to other parts of the human body. It is characterized with a high metastatic potential and a high recurrence rate. For these reasons, the correct and timely diagnosis and treatment is an absolute imperative. To increase the accuracy and speed of urinary bladder cancer diagnosis, in this doctoral dissertation it is proposed to use an intelligent system that will serve as an assistance to medical professionals during diagnosis, treatment and care of patients that suffer with urinary bladder cancer. Methods: To develop such a system, two data sets are collected. The first data set represents the data collected during optical cystoscopy and examination of urinary bladder mucosa with a confocal laser endomicroscope. This data set is used for the development of classification algorithms and it is divided in four classes (healthy mucosa, high-grade carcinoma, low-grade carcinoma and carcinoma in-situ). The other data set used in this research consists of images collected by using computerized tomography (CT). The images collected with CT are captured in three planes (frontal, horizontal and sagittal) and are divided into 6 classes (images without a bladder, healthy bladder, unilateral bladder wall thickening, circular bladder wall thickening, exophytic formation and invasion outside the contour of the bladder). This data set is used during the development of both classification and semantic segmentation algorithms. For the case of semantic segmentation procedure, three different approaches were used. First approach was to use general semantic segmentation system. The second approach was to use a semantic segmentation system that consists of three U-net architectures, one for each plane of CT images. The third approach is to utilize one U-net architecture for each plane and each diagnosis, resulting with 12 separate U-net architectures. With the aim of increasing the performances of developed algorithms, the hybrid models are proposed. The hybrid models are developed by combining standard models based on convolutional neural networks (CNN) with edge detectors, augmentation procedures, meta-heuristic algorithms for model selection or transfer learning paradigm. Results: When the proposed methods are used, it can be noticed that the highest classification results for the case of optical biopsy data set are achieved if a discrete particle swarm (D-PS) algorithm is used to select the model of AlexNet CNN. In this case, micro averaged area under the ROC curve (AUCmicro) over 0.99 is achieved. In the same time, σ(AUCmicro) below 0.005 is achieved. For the case of CT images, the highest classification results are achieved if a ResNet CNN is trained with augmented data set. For the case of images in frontal plane, AUCmicro of 0.99 and σ(AUCmicro) below 0.01 are achieved if ResNet50 is used for classification. In the case of images in horizontal plane, similar results are achieved if ResNet101 network is used. Finally, in the case of sagittal plane, the highest performances are achieved if ResNet152 is used. In this case, AUCmicro of 0.94 and σ(AUCmicro) of 0.05 are achieved. If the semantic segmentation of urinary bladder cancer masses from CT image is observed, it can be seen that the highest performances are achieved if general semantic segmentation system is used. The best-performing model is, in this case, designed with D-PS. In this case dice coefficient (DSC) of 0.99 and σ(DSC) of 0.005 are achieved. It can be noticed that by using separate U-net architectures for each plane, the high-quality results are achieved as well. Conclusions: From the presented results, it can be concluded that, in the case of designed algorithms, high classification and semantic segmentation as well as generalization performances are achieved. It can be noticed that by applying the hybrid approach, an increase of model performances can be achieved. Such a property is particularly emphasized in the case of meta-heuristic algorithms and data augmentation. In final, it can be concluded that there is possibility for the design of an intelligent system for urinary bladder cancer diagnosis that will serve as an assistance tool in the diagnosis, treatment and care of patients that suffer from urinary bladder cancer.