Prostate Cancer Prediction and Detection Using Hybrid CNN-Deep Learning Techniques
Cancer Studies and Therapeutics · 2023
various computer vision tasks, encompassing segmentation, detection, and classification.These techniques leverage convolutional layers to extract distinctive features, progressing from low-level local patterns to high-level global patterns within input images.The incorporation of a fully connected layer at the end of the convolutional layers enables the conversion of intricate patterns into probabilities assigned to specific labels [5,6].The performance of deep learning-based methods can be further enhanced by employing different types of layers, such as batch normalization layers, which normalize layer inputs to possess a zero mean and unit variance, and dropout layers, which randomly exclude selected nodes.Nevertheless, achieving optimal performance necessitates identifying the ideal combination and configuration of these layers, as well as precise tuning of hyperparameters.This challenge persists as one of the primary obstacles in the application of deep learning techniques across diverse domains, including medical imaging.In the context of clinical imagery, each MRI slice contains valuable information pertaining to the location and size of prostate cancer.Ishioka et al. conducted a cut-level analysis involving 318 patients, utilizing U-Net and Res Net models.Remarkably, they achieved an impressive AUC (Area Under the Curve) value of 0.78 on the test set, utilizing only 16 separate slices, without any additional training or validation steps [7,8].The research paper focuses on the following areas: Our study aims to predict the recurrence of prostate cancer using H&E stained tissue AbstractProstate cancer is a widely recognized form of cancer characterized by the proliferation of malignant cells in the prostate gland, a small organ responsible for producing seminal fluid in men.Typically, the progression of prostate cancer is slow and initially confined to the prostate gland, often causing minimal harm.Common symptoms include frequent urination, weak urine flow, blood in the urine or seminal fluid, and pain during urination.However, the current method for detecting prostate cancer, known as Driving While Intoxicated (DWI), suffers from several limitations such as low accuracy, complexity, high computational requirements, and the need for extensive training data.To address these challenges, researchers in the medical imaging field are exploring different Convolutional Neural Networks (CNNs) models and techniques for object detection and segmentation.In this study, a modified CNN system is proposed to develop an automated algorithm capable of detecting clinically significant prostate cancer using DWI images of patients.The study employed a clinical database consisting of 970 DWI images from individuals, with 17 cases diagnosed with prostate cancer (PCa) and 14 cases considered healthy.The performance of the proposed system was evaluated using a training database containing 940 patients, while the remaining 20 patients were reserved for testing.The results demonstrated that the proposed system exhibited improved sensitivity, reduced computational requirements, high performance, and lower time complexity compared to the current prototype system.