ArticlePDF Available

Using classification methodology to improve biomedical decision making

Authors:
Hung-Wen Chiu at Taipei Medical University
Hung-Wen Chiu
Usman Iqbal at UNSW Sydney
Usman Iqbal
Yu-Chuan Jack Li at Taipei Medical University
Yu-Chuan Jack Li
Download full-text PDF
Read full-text
Download citation
Content uploaded by Yu-Chuan Jack Li
Author content
All content in this area was uploaded by Yu-Chuan Jack Li on Aug 07, 2019
Content may be subject to copyright.
computer methods and programs in biomedicine 137 (2016) A1–A2
journal homepage: www.intl.elsevierhealth.com/journals/cmpb
Editorial
Using classication methodology to improve
biomedical decision making
Classication methodology has been widely used in biomedi-
cal data processing to support decision making. The new
algorithms and new applications are developed to facilitate
the basic research and clinical practice. In this month, the
editors’ choice articles demonstrate different classication
methods to solve problems in different biomedical domains.
The rst editors’ choice, “Detection of exudates in fundus
photographs using deep neural networks and anatomical
landmark detection fusion [1],” presents a deep convolutional
neural networks for early exudates detection of diabetic reti-
nopathy. In this research, deep neural networks were used as
a classier to generate exudates probability map cooperating
other image processing procedures for detection of exudates
in fundus photographs. The image processing procedures
included vessel detection and optic disc detection to increase
the accuracy. There are four convolutional layers and four
max-pooling layers in this deep neural network architecture.
The results revealed that this proposed method can achieve
a performance with sensitivity 0.78 and specicity 0.78 that
is better than previous researches. The authors suggested
that automated segmentation using convolutional neural
networks in combination with other landmark detectors is an
important step in creating automated screening programs for
early detection of diabetic retinopathy.
Chen et al. present another work classication problem in
automatic medical image diagnosis titled as “Examining pal-
pebral conjunctiva for anemia assessment with image pro-
cessing methods [2].” In this study the authors propose the
algorithms for anemia diagnosis. In addition to a simple and
fast method with image processing, a more sophisticated and
robust method with machine-learning based classication
methods were also proposed. The results showed that using
support vector machine and articial neural network with the
three selected features to do classication can achieve a
better recognition performance. For example, the sensitive
and specicity are 0.78 and 0.83 in the robust algorithm with
SVM using 100 testing samples with 10-fold cross validation.
This study shows the feasibility of the proposed classication
method and gives a promising direction for the follow-up
study in clinical application.
For basic biomedical research, classication-based method
also shows its potential. Kim et al. propose a new classica-
tion-based method for stable isotope labeling by amino acids
in cell culture (SILAC) data [3]. Particle swarm optimization
(PSO) classication method was applied in this study to iden-
tify the differentially abundant proteins that are statistically
either down-regulated or up-regulated. The authors used
simulation data and experiment data to verify the perfor-
mance of the proposed methods and compare to other
methods. The simulation studies show that the newly devel-
oped PSO-based method performs the best among others in
terms of F1 score. Furthermore the proposed methods dem-
onstrate the ability of detecting potential markers through
real SILAC experimental data.
The above editors’ choice articles incorporated classication
method in detection of exudates in fundus photographs,
examining palpebral conjunctiva for anemia assessment,
and stable isotope labeling by amino acids in cell culture
data. Many new algorithms are proposed and many domain
applications are presented and compared. However, how to
select correct classication techniques to improve the per-
formance of medical decision support system remains
exploration.
REFERENCES
[1] P. Prentašić, S. Lončarić, Detection of exudates in fundus
photographs using deep neural networks and anatomical
landmark detection fusion, Comput. Methods Programs
Biomed. 137 (2016) 281–292.
[2] Y.-M. Chen, S.-G. Miaou, H. Bian, Examining palpebral
conjunctiva for anemia assessment with image processing
methods, Comput. Methods Programs Biomed. 137 (2016)
125–135.
[3] S. Kim, N. Carruthers, J. Lee, S. Chinni, P. Stemmer,
Classication-based quantitative analysis of stable isotope
labeling by amino acids in cell culture (SILAC) data, Comput.
Methods Programs Biomed. 137 (2016) 137–148.
Downloaded for Anonymous User (n/a) at Taipei Medical University from ClinicalKey.com by Elsevier on October 23, 2017.
For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.
A2        () A–A
Hung-Wen Chiua
aGraduate Institute of Biomedical Informatics, College of Medicine
Science and Technology, Taipei Medical University, Taipei, Taiwan
Usman Iqbalb,c
bMaster Program in Global Health and Development, College of
Public Health, Taipei Medical University, Taipei, Taiwan
cInternational Center for Health Information Technology (ICHIT),
Taipei Medical University, Taipei, Taiwan
Yu-Chuan (Jack) Lia,c,d,*
aGraduate Institute of Biomedical Informatics, College of Medicine
Science and Technology, Taipei Medical University, Taipei, Taiwan
cInternational Center for Health Information Technology (ICHIT),
Taipei Medical University, Taipei, Taiwan
dChair, Dept. of Dermatology, Wan Fang Hospital, Taipei, Taiwan
*Corresponding author at: 250- Wuxing Street, Xinyi District,
Taipei 11031, Taiwan. Fax: +886 2 6638 7537.
E-mail address: jack@tmu.edu.tw; jaak88@gmail.com
(Y.-C. Li).
Downloaded for Anonymous User (n/a) at Taipei Medical University from ClinicalKey.com by Elsevier on October 23, 2017.
For personal use only. No other uses without permission. Copyright ©2017. Elsevier Inc. All rights reserved.
ResearchGate has not been able to resolve any citations for this publication.
Detection of exudates in fundus photographs using deep neural networks and anatomical landmark detection fusion
Article
  • Oct 2016
Background and objective: Diabetic retinopathy is one of the leading disabling chronic diseases and one of the leading causes of preventable blindness in developed world. Early diagnosis of diabetic retinopathy enables timely treatment and in order to achieve it a major effort will have to be invested into automated population screening programs. Detection of exudates in color fundus photographs is very important for early diagnosis of diabetic retinopathy. Methods: We use deep convolutional neural networks for exudate detection. In order to incorporate high level anatomical knowledge about potential exudate locations, output of the convolutional neural network is combined with the output of the optic disc detection and vessel detection procedures. Results: In the validation step using a manually segmented image database we obtain a maximum F1 measure of 0.78. Conclusions: As manually segmenting and counting exudate areas is a tedious task, having a reliable automated output, such as automated segmentation using convolutional neural networks in combination with other landmark detectors, is an important step in creating automated screening programs for early detection of diabetic retinopathy.
View
Classification-based quantitative analysis of stable isotope labeling by amino acids in cell culture (SILAC) data
Article
  • Sep 2016
Background and objective: Stable isotope labeling by amino acids in cell culture (SILAC) is a practical and powerful approach for quantitative proteomic analysis. A key advantage of SILAC is the ability to simultaneously detect the isotopically labeled peptides in a single instrument run and so guarantee relative quantitation for a large number of peptides without introducing any variation caused by separate experiment. However, there are a few approaches available to assessing protein ratios and none of the existing algorithms pays considerable attention to the proteins having only one peptide hit. Methods: We introduce new quantitative approaches to dealing with SILAC protein-level summary using classification-based methodologies, such as Gaussian mixture models with EM algorithms and its Bayesian approach as well as K-means clustering. In addition, a new approach is developed using Gaussian mixture model and a stochastic, metaheuristic global optimization algorithm, particle swarm optimization (PSO), to avoid either a premature convergence or being stuck in a local optimum. Results: Our simulation studies show that the newly developed PSO-based method performs the best among others in terms of F1 score and the proposed methods further demonstrate the ability of detecting potential markers through real SILAC experimental data. Conclusions: No matter how many peptide hits the protein has, the developed approach can be applicable, rescuing many proteins doomed to removal. Furthermore, no additional correction for multiple comparisons is necessary for the developed methods, enabling direct interpretation of the analysis outcomes.
View
Examining palpebral conjunctiva for anemia assessment with image processing methods
Article
  • Sep 2016
  • COMPUT METH PROG BIO
Examining the hemoglobin level of blood is an important way to achieve the diagnosis of anemia, but it requires blood drawing and blood test. Examining the color distribution of palpebral conjunctiva is a standard procedure of anemia diagnosis, which requires no blood test. However, since color perception is not always consistent among different people, we attempt to imitate the way of physical examination of palpebral conjunctiva to detect anemia, so that computers can identify anemia patients automatically in a consolidated manner for a screening process. In this paper we propose two algorithms for anemia diagnosis. The first algorithm is intended to be simple and fast, while the second one to be more sophisticated and robust, providing an option for different applications. The first algorithm consists of a simple two-stage classifier. In the first stage, we use a thresholding decision technique based on a feature called high hue rate (HHR) (extracted from the HSI color space). In the second stage, a feature called pixel value in the middle (PVM) (extracted from the RGB color space) is proposed, followed by the use of a minimum distance classifier based on Mahalanobis distance. In the second algorithm, we consider 18 possible features, including a newly added entropy feature, some improved features from the first algorithm, and 13 features proposed in a previous work. We use correlation and simple statistics to select 3 relatively independent features (entropy, binarization of HHR, and PVM of G component) for classification using a support vector machine or an artificial neural network. Finally, we evaluate the classification performance of the proposed algorithms in terms of sensitivity, specificity, and Kappa value. The experimental results show relatively good performance and prove the feasibility of our attempt, which may encourage more follow-up study in the future.
View