Cervical precancerous lesions classification using pre-trained densely connected convolutional networks with colposcopy images. (January 2020)

• Record Type: Journal Article
• Authors: Zhang, Tao; Luo, Yan-min; Li, Ping; Liu, Pei-zhong; Du, Yong-zhao; Sun, Pengming; Dong, BinHua; Xue, Huifeng
• Is Part Of: Biomedical signal processing and control. Volume 55(2020)
• Publication Date: 2020-01
• Subjects: Colposcopy -- Cervical precancerous -- Computer-aided diagnosis -- Densely connected convolutional networks -- Transfer learning; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2019.101566
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: This study is the first to apply the transfer learning with the DenseNet-based model to the classification of cervical colposcopy images and achieved satisfactory results for computer aided diagnosis. Experiments fine-tuned limited cervical images with different pre-trained DenseNet model (ImageNet and Kaggle) and the relative performance and causes of different fine-tuned models are analyzed. This study conducted enough comparative experiments. Compared with traditional machine learning methods, this study proves that deep learning is more suitable for cervical colposcopy image classification. Comparative experiments on model training show that deep learning is more inclined to transfer learning and more training data in the analysis of limited medical images. This paper lists all previous studies on deep learning in cervical colposcopy image analysis. In comparison, the proposed method achieves very satisfactory results in image-only training. Importantly, comparisons with the clinical group indicate that our approach is comparable to that of senior physicians. Although the clinical group is limited, this study is meaningful for the development of the computer-aided diagnosis of cervical lesion screening. Abstract: Colposcopy is currently a common medical technique for preventing cervical cancer. However, with the increase of the workload, screening by artificial vision has the problems of misdiagnosis and low diagnostic efficiency. Based on transfer learning, pre-trained densely connected convolutional networks are used to propose a computer-aided-diagnosis (CAD) method for automatic classification of cervical precancerous. The proposed method is applied to determine CIN2 or higher-level lesions in cervical images. In the present work, image data are initially prepossessing with ROI extraction and data augmentation. Then, parameters of all layers are fine-tuning with pre-trained DenseNet convolutional neural networks from two datasets (ImageNet and Kaggle). The impact of different training strategies on the model performance with limited training data is analyzed, including random initialization (RI) training from scratch, fine-tuning (FT) pre-trained model, different size of training data and K-fold cross-validation. Experimental results show that our method (FT) achieves an accuracy of 73.08% (AUC ≈ 0.75) in 600 test images. Compared with previous related work and clinicians, the performance of our approach can effective diagnosis CIN2+ and comparable with a senior physician, which proves the feasibility and promising of the proposed computer-aided diagnostic method.