An Adaptive Regularization Approach to Colonoscopic Polyp Detection Using a Cascaded Structure of Encoder–Decoders

Maxwell Hwang,Da Wang,Wei-Cheng Jiang,Xiang Pan,Dongliang Fu,Kao-Shing Hwang,Kefeng Ding
DOI: https://doi.org/10.1007/s40815-019-00694-y
IF: 4.085
2019-01-01
International Journal of Fuzzy Systems
Abstract:This research aims to segment colonoscopic images by automatically extracting polyp features by exploiting the strengths of convolution neural networks (CNN). The proposed model employs deep learning and adaptive regularization techniques. The model is structurally composed of two cascaded encoder–decoder networks, each of which is constructed by four CNN layers and two full connection layers. The front model is built on backpropagation learning for segmenting a colonoscopic polyp image. The output images from the precedent hetero-encoder are regarded as corrupted labeled images, especially during the time period close to the end of learning, and are selectively fed into the successive auto-encoder for denoising learning to enhance its discriminative power and relieve the problem of a lack of labeled data for medical image tasks. The performance of the proposed model can be further improved by a simple fuzzy logic approach setting the regularization parameter in the loss function. The proposed method utilizes features learned from some open medical datasets and our own collected dataset. The performance of the proposed architecture is compared with a state-of-the-art network. The evaluation shows the performances of the proposed method are consistent across all the datasets and often outperform the state-of-art model.
What problem does this paper attempt to address?