A Spatial Regulated Patch-Wise Approach for Cervical Dysplasia Diagnosis

Authors: Ying Zhang, Yifang Yin, Zhenguang Liu, Roger Zimmermann733-740

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on an 18-year real-world dataset indicate a minimal of 3.47%, 4.59%, 8.54% improvements over the state-of-the-art in accuracy, F1, and recall measures, respectively.
Researcher Affiliation Academia Ying Zhang1,2,Yifang Yin1*, Zhenguang Liu3*, Roger Zimmermann1 1School of Computing, National University of Singapore 2School of Computer Science, Northwestern Polytechnical University 3School of Computer and Information Engineering, Zhejiang Gongshang University {dcszyi, idsyin, dcsrz}@nus.edu.sg, liuzhenguang2008@gmail.com
Pseudocode Yes Algorithm 1: Iterative spatial regulator update. Result: Regulators S Initialization: S0 i = G0; for each training epoch t do for each training image Ii do {f(xj i)} Patch feature extraction; {aj i} Patch weight learning; (µ, σ) Parameters estimation from patch weights; if (µ, σ) is valid then ˆSt(Ii) Update the regulator; end end
Open Source Code No The paper does not provide any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets No We evaluate the approach on a real-world database from the U.S. National Cancer Institute (NCI). The dataset is accessible based on request and under constrained agreement.
Dataset Splits No 80% randomly selected data is for training while the remaining 20% is for testing. The paper does not explicitly mention a validation split.
Hardware Specification No The paper does not provide specific details about the hardware used to run its experiments.
Software Dependencies No The paper describes network architectures and activation functions (e.g., CNN, Re Lu, Sigmoid) but does not specify any software libraries or frameworks with version numbers (e.g., Python, TensorFlow, PyTorch versions) used for implementation.
Experiment Setup Yes Parameters By default, all images are resized to same of 42x42 patches with patch size 28. The feature representation CNN has three convolutional layers with 12, 24, 48 filters and the size of filter is 3 3, followed by Re Lu activation and max-pooing. The weights-learning module contains three dense layers of 800, 512, 128 nodes. The aggregated feature is passed to a dense layer activated by Sigmoid. The default parameters are Xavier initialized for all layers. By default, the learning rate is 10 5 and λ is 0.1 over a 150-epoch training with early-stop mechanism.