File(s) not publicly available
Classification and Detection of Cancer in Histopathologic Scans of Lymph Node Sections Using Convolutional Neural Network
journal contributionposted on 2023-07-26, 15:54 authored by Misbah Ahmad, Imran Ahmed, Messaoud A. Ouameur, Gwanggil Jeon
Cancer has been considered one of the major threats to the lives and health of people. The substantial clinical practices show that earlier diagnosis and detection of cancer can provide adaptable treatment methods, increase survivability, and enhance life quality. Moreover, rapid advancements in science, technology, and Computer-Aided Diagnosis systems also provide additional information for robust analysis and examination of medical images. Image processing and machine learning presented promising low-cost approaches for classifying and detecting different cancerous diseases. However, these traditional techniques need extensive pre-processing and laborious manual features extraction methods. Thus, in this paper, we presented a Convolutional Neural Network based method for the classification and detection of metastatic cancer in histopathologic images of lymph node sections. A diagnostic method of cancer in histopathologic images is time consuming and tedious for pathologists because a large tissue area has been examined, and tiny metastasis can be easily ignored. Thus the developed deep learning method can help pathologists in examining the histopathologic scans and assist in decision-making to analyze the disease and cancer staging, which will give consequential opinions in clinical diagnosis. We performed the necessary pre-processing and data augmentation steps to enhance the results and avoid overfitting. The method utilizes low dimensional representations and performs automated, categorical feature extraction and classification, which attain high accuracy for diagnosis of cancer. The method is applied to PatchCamelyon (PCam) data set. Experimental results show good performance with an accuracy rate of 0.94 for the medical image classification and detection task.
Publication titleNeural Processing Letters