2026/01/26 by Dilber U. Ozsahin, Declan Ikechukwu Emegano, Mubarak T. Mustapha +2 · 1 voice
Computer Science · Medicine · #AI in cancer detection #COVID-19 diagnosis using AI #Prostate Cancer Diagnosis and Treatment
paper · doi:10.17159/sajs.2026/18193
openalex publication_date 2026/01/26 · openalex created_date 2026/01/27 · openalex updated_date 2026/07/22
Prostate cancer is the most common solid tumour in men and the fifth leading cause of cancer death globally. It requires timely and accurate diagnostic procedures for the treatment processes. However, these procedures are labour intensive because of the histological examination of prostate biopsy specimens, which can be subject to interpretative variability. The present study was designed to evaluate the effectiveness of deep-learning algorithms specifically for the task of classifying prostate biopsy images into two categories: benign or malignant. The data set included 247 cancerous and 514 benign histological biopsy images. The data set was derived from patients aged between 39 and 80 years and who underwent prostate biopsies at the Federal Teaching Hospital in Lokoja, Nigeria, between 2019 and 2023. We augmented the data set to 10 000 histological images, after which 50 images from the same cohort were reserved for validation. Multiple Source Hierarchical Aggregation Neural Network, densely connected convolutional network, EfficientNet, Inception v3, MobileNet, ResNet-50, Visual Graphics Group 16 and Visual Graphics Group 19 were among the deep-learning models that were trained and verified. The results showed that densely connected convolutional network had an accuracy value of 0.96, with precision, recall and F1 scores of 1.00, 0.92 and 0.96, respectively, for benign cases and 0.93, 1.00 and 0.96, respectively, for malignant cases. Deep-learning models, especially the densely connected convolutional network, have shown great potential to distinguish between benign and malignant prostate images; as a result, they can significantly enhance prostate cancer diagnosis by improving diagnostic uniformity and efficacy for pathologists.