# Research Project Title: Machine-to

```Research Project Title: Machine-to-Machine Application of Telehealth Using Automatic
Electrocardiographic Analysis System by Digital Signal Processing Methods
1. Team member
Principal Investigators:
Co-PI:
Assoc Prof. Yi-Lwun Ho
Assist Prof. Jian-Jiun Ding; Assist Prof. Yen-Hung Lin; Dr. Ying-Hsien Chen
2. Discussion with champions:
Time: 6/29 8:00-9:00
We also discussed with other PIs and Co-PIs at the meeting on 6/16. After discussion, we found that
there are many possible ways to further extension the E-health system.
3. Progress between last month and this month (the algorithm for extracting the feature points and
differentiation of normal and abnormal ECG including APC, VPC, atrial fibrillation)
a. Summary of key findings and innovation, and/or
 The results of the previous months:
For the MIT/BIH Arrhythmia Database (used worldwide)
The number of R points:
True Positive (TP):
110159,
109922
False Negative (FN): 237
False Positive (FP): 68
Recall rate = TP/(TP+FN) = 99.78%
Precision rate = TP/(TP + FP) = 99.94%
Error rate = (FN+FP)/(TP+FN) = 0.2769%
Computation time:
0.6 second for a 30-minute length ECG signal (by Matlab).
Comparison: For the best result in the literature, error rate = 0.5516%
Although recently there was a paper that presents a result with the error rate of
0.35%, we doubt that the authors did not apply the standard ground truth.
Stage
MULS
Baseline Extraction
8N
11N
R Point Candidates (I)
3N
7N
R Point Candidates (II)
2N1
4N1
R Point Determination
3N2
8N2
SUM
LOGICS
SUM
19N
2N
12N
6N1
15N2
26N2
31N + 6N1 + 26N2
N1  N/5, N2  N/20, and the number of instructions is near to 33.5N, where N is the number of
samples
If N = 100 per second, then the energy consumption is 33.5 W.
 The results of this month:
(1) We have developed some possible ways to further reduce the number instructions. For example,
the baseline extraction algorithm can be much simplified and the total number of instructions can be
reduced to about 20N. But in this case the error may increase to 1.5%.
(2) We write our work about R point extraction as a conference paper. We plan to submit it into
National Symposium on Telecommunications. The conference paper has been completed.
(3) We also write our work about ECG signal analysis as an SCI journal paper. We plan to submit it
into IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING. The paper is in process (10 pages
have been written).
b. Cross-project synergy
After discussion with the PIs and Co-PIs of other projects, we found that there are a lot of ways
to further extend the E-health system. For example, the proposed ECG analysis system can be
integrated with the systems for analyzing the pupil size, the eye ground, the ear lobe crease, blood
pressure, blood sugar, and the amount urine protein for vasculopathy diagnosis. This idea is a possible
way for future research if we can cooperate with the faculties who are familiar with image processing
and machine learning.
4. Brief plan for the next month
(1) We will submit the conference paper and complete the SCI journal paper about R point extraction.
(2) We will spend time to study machine learning. It is important to analyze the symptom of a patient.
a. Updated view of planned milestones, deliverables, and success criteria
Date
Milestone
Deliverables
Technical Success
Criteria/ Objectives
2012Q2
Make solid power
 Presentation of the study
 Confirm the research
direction
first risk of this project
2012Q4
Finish the first version of  Technical report
 A technique report
the ECG interpretation
 Software for the interpretation
with detailed
algorithm.
 Define some joint develop
description
items with other subproject.  Clinical trial IRB
application
2012Q5
Develop a very accurate  Technical report

For MIT/BIH
and very fast algorithm for  Software for the interpretation Arrhythmia Database,
ECG feature point
 Define some joint develop
sensitivity = 99.80% and
extraction.
items with other subproject.
computation time for each
 Discuss the arrhythmia
30 minute length ECG
problem
data is 0.3 second.
2012Q6
Develop the way to much  Technical report of the system 
The number of
reduce the number of
analysis result
instructions for the
instructions for ECG signal
proposed system is
analysis
reduced to 33.5N
2012Q7
Propose the idea for further  Complete a conference paper. 
The number of
extend the H-health system  Start to write an SCI journal instructions can be further
such as vasculopathy
paper about R-point extraction reduced to 20N and the
diagnosis
system still work well for
normal cases.
5. Research byproducts
(1) Modify the theory about the bilateral filter. (2) Propose the geometric coefficient filter, which
requires very less number of instructions. In addition to ECG signal analysis, the modified bilateral
filter and the geometric coefficient filter can also be applied in other fields of signal processing.
```