Classification of Prostate Precancerous Cells Using DenseNet-201 and Inception-v3 Models

Prostate cancer is globally recognized as the fourth most common cancer, with approximately 1.3 million cases reported annually. Diagnosing prostate cancer remains challenging due to the potential for diagnostic errors caused by unclear imaging, leading to prolonged diagnostic times. To address this issue, artificial neural networks offer a promising solution by assisting medical teams in classifying prostate cells with greater speed and accuracy. This study focuses on the comparative analysis of two convolutional neural network (CNN) models, DenseNet-201 and Inception-v3, for prostate cell classification. DenseNet-201 demonstrated superior accuracy during training, achieving an average of 98.85% (STD ± 0.89), compared to Inception-v3's 98.49% (STD ± 0.85). However, Inception-v3 excelled in processing efficiency, with an average runtime of 19 minutes and 6 seconds, significantly outperforming DenseNet-201's 95-minute runtime. During testing, both models achieved exceptional classification performance, with DenseNet-201 and Inception-v3 recording average accuracies of 97.32% and 97.39%, respectively. These findings confirm the effectiveness of both models for prostate cell classification, with DenseNet-201 excelling in accuracy and Inception-v3 offering faster processing. This research highlights the potential of CNNs in improving diagnostic workflows and underscores their applicability in medical image analysis. Future studies may explore hybrid models or optimizations to balance accuracy and efficiency for clinical deployment.