Olawuyi Olushola, Viriri Serestina
School of Mathematics, Statistics and Computer Science, University of KwaZulu-Natal, Durban 3629, South Africa.
J Imaging. 2025 Jul 28;11(8):254. doi: 10.3390/jimaging11080254.
The human interpretation of medical images, especially for the detection of cancer in the prostate, has traditionally been a time-consuming and challenging process. Manual examination for the detection of prostate cancer is not only time-consuming but also prone to errors, carrying the risk of an excess biopsy due to the inherent limitations of human visual interpretation. With the technical advancements and rapid growth of computer resources, machine learning (ML) and deep learning (DL) models have been experimentally used for medical image analysis, particularly in lesion detection. However, several state-of-the-art models have shown promising results. There are still challenges when analysing prostate lesion images due to the distinctive and complex nature of medical images. This study offers an elaborate review of the techniques that are used to diagnose prostate cancer using medical images. The goal is to provide a comprehensive and valuable resource that helps researchers develop accurate and autonomous models for effectively detecting prostate cancer. This paper is structured as follows: First, we outline the issues with prostate lesion detection. We then review the methods for analysing prostate lesion images and classification approaches. We then examine convolutional neural network (CNN) architectures and explore their applications in deep learning (DL) for image-based prostate cancer diagnosis. Finally, we provide an overview of prostate cancer datasets and evaluation metrics in deep learning. In conclusion, this review analyses key findings, highlights the challenges in prostate lesion detection, and evaluates the effectiveness and limitations of current deep learning techniques.
传统上，人类对医学图像的解读，尤其是前列腺癌检测，一直是一个耗时且具有挑战性的过程。手动检查前列腺癌不仅耗时，而且容易出错，由于人类视觉解读的固有局限性，存在过度活检的风险。随着技术进步和计算机资源的快速增长，机器学习（ML）和深度学习（DL）模型已被用于医学图像分析的实验，特别是在病变检测方面。然而，一些最先进的模型已显示出有前景的结果。由于医学图像独特而复杂的性质，在分析前列腺病变图像时仍存在挑战。本研究对使用医学图像诊断前列腺癌的技术进行了详尽综述。目标是提供一个全面且有价值的资源，帮助研究人员开发准确且自主的模型，以有效检测前列腺癌。本文结构如下：首先，我们概述前列腺病变检测的问题。然后我们回顾分析前列腺病变图像的方法和分类方法。接着我们研究卷积神经网络（CNN）架构，并探索其在基于图像的前列腺癌诊断深度学习（DL）中的应用。最后，我们概述深度学习中的前列腺癌数据集和评估指标。总之，本综述分析了关键发现，突出了前列腺病变检测中的挑战，并评估了当前深度学习技术的有效性和局限性。
