This thesis explores the development and evaluation of deep learning models for classifying prostate cancer using 3D-based Multiparametric Magnetic Resonance Imaging (mpMRI). The study compares the performance of pre-trained models (ResNet50, EfficientNet and MedicalNet) with a custom built Convolutional Neu- ral Network (CNN) model trained from scratch. The methods include data prepro- cessing, model training and evaluation using standard metrics such as sensitivity, specificity, F1-Score, and AUC. The results show that MedicalNet, a pre-trained model designed for medical imaging, achieve the best balance between sensitiv- ity (66.4%) and specificity (85.31%), while the custom CNN model exhibits the highest sensitivity (84.4%), making it the most effective at detecting cancer cases. In contrast, the EfficientNet model has the lowest accuracy, emphasizing the im- portance of domain specific pre-training. This study demonstrates the potential of state-of-the-art deep learning models in improving prostate cancer diagnosis, indicating that well chosen pre-trained models can provide strong diagnostic per- formance. Future work should focus on enhancing class imbalance management and testing the models on larger and more diverse datasets.