Fine-Tuned Deep Learning Architectures for Accurate Prostate Cancer Detection

Prostate cancer is one of the most common diseases that affects men after the age of 60 years. Still, nowadays men after 40 years have to take several prevention methods to predict and recognise prostate cancer in the early stages. Many existing approaches have already been developed and used to detect prostate cancer using Prostate CT scan images. This paper presents the Fine-Tuned Learning Architecture (FTLA) approach, which contains multiple models that detect and classify normal and abnormal regions in prostate CT scan images. The pre-trained model, Multi-Layered ResNet101 with transfer learning, plays a significant role in identifying the accurate patterns of Prostate cancer regions. To improve the input CT scan image in terms of contrast and smoothing, firstly Contrast-Limited Adaptive Histogram Equalization (CLAHE) and ADF filtering is applied to obtain the refined image, followed by feature extraction techniques such as Histogram of Oriented Gradients (HOG) and Variational Autoencoders (VAE) extracts the edge, shape and latent features which shows huge impact on classification of prostate cancer. In the final stage, the refined model, Multi-Attention U-Net (MA-UNet), was used and integrated with the Fine-Tuned Multi-Layered Classifier (FTMLC) to classify accurate CT scan images. Researchers consider two datasets: one from Kaggle, collected from online sources, and the second, real-time CT scan images. The results show that the proposed approach achieves superior performance.