�ݺ�ߣ

Medical Analytics As a
Service (MaaS)
API
RESTful API for seamless
integration. Historical and
ongoing health data accessible in
seconds to minutes.
Analytics
Comprehensive health data
analytics powered by machine
learning. Our platform becomes
smarter with use.
Medical Database
Our medical knowledge base
composed of verified knowledge,
statistical facts, and formulas
curated by a team of data
scientists.

Medical Analytics As a
Service (MaaS)
Our services
API System
Chromosome
Karyotyping
ECG
Analysis
Medtrics
MaaS

MaaS ECG Analysis
Recognize 90% common heart disease ECG
Atrial Fibrillation
Irregular and often rapid heart rate that can increase your risk of
stroke, heart failure and other heart-related complications.
Premature ventricular contraction (PACs)
Premature arterial contractions, also known as atrial premature
complexes (APC) or atrial premature beats (APB), are a common
cardiac dysrhythmia characterized by premature heartbeats
originating in the atria.

MaaS ECG Analysis
Recognize 90% common heart disease ECG
Premature Ventricular Contractions��
(PVCs) Fibrillation
Premature ventricular contractions (PVCs) are extra, abnormal
heartbeats that begin in one of your heart’s two lower pumping
chambers (ventricles). These extra beats disrupt your regular heart
rhythm, sometimes causing you to feel a flip-flop or skipped beat in
your chest.
Sinus Bradycardia
Sinus bradycardia is a sinus rhythm with a rate that is lower than
normal. In humans,bradycardia is generally defined to be a rate of
under 60 beats per minute.

MaaS ECG Analysis
Advantages
Validated against MIT, AHA Databases and
more…
On par or better than Stanford ML Group Arrhythmia
Detection model
http://stanfordmlgroup.github.io/projects/ecg/
Comprehensive heart disease detection (AFib,
PAC, PVC…)
Works with all 12 leads

MaaS ECG Analysis
AFIB Algorithm
Lorenz Plots
Based on Lorenz plots: 2D plot of the
evolution of the difference between
consecutive beats intervals (RR intervals).
Atrial fibrillation on figure (f).

MaaS ECG Analysis
AFIB Tests Results
Positive + Negative -
Positive
+
True Positive (TP)
AFIB detected; window with AFIB.
False Positive (FP)
AFIB detected; window with no
AFIB.
Positive Predictive Value
(PPV) = TP/(TP+FP)
Negative
-
False Negative (TN)
No AFIB detected; window with >
60% AFIB
True Negative (TN)
No AFIB detected; window with
no or < 60% of AFIB
Sensitivity = TP/(TP+FN)
Analysis Outcome
MITAnnotation
Windows of 60s against MIT-BIH AFIB db23 annotated records; ~10hr long each.

MaaS ECG Analysis
AFIB Tests Results & Statistics Description
AF detectors comparison table (reference: “IMPLEMENTATION AND TESTING OF ATRIAL FIBRILLATION DETECTORS FOR A
MOBILE PHONE APPLICATION”, Roberta Colloca)
Algorithm Database
Performance
WL
Se(%) PPV(%)
Medtrics Algorithm MIT-BIH AF db 94.7 96.7 60s
Lake and Moorman MIT-BIH AF db 91.0 - 12bs
Cerutti et al. MIT-BIH AF db 96.1 75.76 90s
Linker DT. MIT-BIH AF db 97.64 81.81 10s
Tateno et al. MIT-BIH AF db 91.2 90.32 50s
Sarkar et al. MIT-BIH AF db 97.5 99.0 120s
Logan et al. MIT-BIH AF db 87.3 85.72 120s
Petrucci et al. 5 MIT-BIH AF db 92 78 60s
Petrucci et al. 6 MIT-BIH AF db 91 92 40s
Babaeizadeh et al. MIT-BIH AF db 87.27 92.75 40s

MaaS ECG Analysis
AFIB Algorithm Output Example - Excerpt of record 04043
AFIB evidence output of the
detection algorithm:
● > 20: AFIB detected
● < 20: No AFIB detection
● When different from 0, it
indicates analyzed 60s
window having >60% of
AFIB (as annotated in the
MIT database)

MaaS ECG Analysis
MaaS Portal -ECG processing

MaaS ECG Analysis
Denoise ECG metrics

MaaS ECG Analysis
Body Stress signals from ECG metrics (HRV)

AFib detection
Isolated PAC, PVC
HRV Analysis
ECG De-noise and baseline removal
QRS, QT, PR, and ST deviation intervals
Wide QRS complex, long/short QTc interval
AFib plot
ECG Template
Sinus Bradycardia
Hyper(po)calcaemia/Hyper(po)kalaemia signs
Monomorphic VT on sequences of beats
MaaS ECG Analysis
Roadmap
Phase I
PVC bigeminy, trigeminy, quadrigeminy
(PVCs with normal beat in between)
PAC bigeminy, trigeminy, quadrigeminy
(PACs with normal beat in between)
PVC Couplet/Triplet (Consecutive PVCs)
PAC Couplet/Triplet (Consecutive PACs)
PAC with abnormal morphologies
Beat by Beat VT
Phase II

MaaS Karyotyping
with Neural Network
Current Issue
Medtrics
Progress
Time consumed in
karyotyping test
2 hours manual work by technician
for each karyotyping test.
Potential error
Potential human error
High cost
~200USD per test in Taiwan
~1,000USD per test in USA

MaaS Karyotyping
with Neural Network
Our Solution
Machine Learning
We automate chromosome karyotyping by deep
learning and Computed Vision technologies.
Lab Test Efficiency
Increase the efficiency of doing lab test.
After
Before

MaaS Karyotyping
with Neural Network
Advantages
Automated solution. Each karyotyping test will
take 30 secs.
Proven 99.7% classification accuracy.
Eliminate human error.
Increase sample size examination: can examine all
100 cells within 1 blood/amniotic fluid sample vs
1 cell. Increase total confidence level like
Mosaicism Exclusion.
Much lower cost.

MaaS Karyotyping
with Neural Network
Current Status
Isolated chromosome
classification: 99.7% accuracy
Overlapped chromosome
extraction algorithm in progress.
Target for trial in 2 months

MaaS Karyotyping
with Neural Network
Roadmap
Isolated chromosome classificationPhase I
Create mask over chromosome in raw cell clusters (~20
from amniotic fluid and ~100 from blood sample)
Identify and separate overlapping chromosomes
Abnormal chromosome detection
• Abnormal chromosome segments (Mosaicism)
• Extra/Missing chromosome
Lab test workflow integration
Phase II

MaaS Interactive API
Target for health IT developer

MaaS Business Model
Market Strategy
PHASE I
PHASE II
PHASE III
Device Sign Up
Scale up Subscriber
Medical Professional
Platform

�ݺ�ߣ

Maa s bisa

Recommended

More Related Content

Similar to Maa s bisa (20)

Recently uploaded (20)

Maa s bisa