24
1 The e-Logistics of The e-Logistics of Securing Securing Distributed Medical Data Distributed Medical Data Andrew M. Snyder Andrew M. Snyder Alfred C. Weaver Alfred C. Weaver

1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

Embed Size (px)

Citation preview

Page 1: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

11

The e-Logistics of Securing The e-Logistics of Securing Distributed Medical DataDistributed Medical Data

Andrew M. SnyderAndrew M. Snyder

Alfred C. WeaverAlfred C. Weaver

Page 2: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

22

Medical Data Portal Web Services

AuthorizationService

AuthenticationService

Electronic Patient Record

2

3

9

10

11

12

RuleEngines

1

46

7

5

8

Medical Data AccessMedical Data Access

Page 3: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

33

IssueIssueHIPAA requires that electronic medical data HIPAA requires that electronic medical data by encrypted when stored or transmittedby encrypted when stored or transmitted

This is not an issue for a single x-rayThis is not an issue for a single x-ray

But U. Virginia radiology does 380,000 But U. Virginia radiology does 380,000 examinations per year and generates 9 TB of examinations per year and generates 9 TB of data annuallydata annually

What is the workflow impact of encrypting What is the workflow impact of encrypting and decrypting data (especially images) and decrypting data (especially images) every time they are touched?every time they are touched?

Page 4: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

44

Encryption IssuesEncryption Issues

Symmetric keySymmetric key– DES, 3DES, AES, othersDES, 3DES, AES, others

Public keyPublic key– RSARSA

Key lengthKey length

Key managementKey management

Managed vs. unmanaged codeManaged vs. unmanaged code

Workflow impactWorkflow impact

Page 5: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

55

Managed vs. Unmanaged CodeManaged vs. Unmanaged Code

Unmanaged codeUnmanaged codenative codenative code

optimized for a device/platformoptimized for a device/platform

advantage: fastadvantage: fast

Managed codeManaged codeexecuted inside a containerexecuted inside a container

translated at runtimetranslated at runtime

provides memory managementprovides memory management

provides garbage collectionprovides garbage collection

advantages: safe, secure, portableadvantages: safe, secure, portable

Page 6: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

66

Rationale for New MeasurementsRationale for New Measurements

No published body of performance No published body of performance measurements for .NET cryptographic servicesmeasurements for .NET cryptographic services

No published understanding of the costs of No published understanding of the costs of managed code (e.g., C#, Java)managed code (e.g., C#, Java)

No insight into how HIPAA's encryption No insight into how HIPAA's encryption requirement will impact an academic radiology requirement will impact an academic radiology departmentdepartment

Page 7: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

77

Performance MeasurementsPerformance Measurements

TestbedTestbed– ComputerComputer

Visual Studio .NET 2003Visual Studio .NET 2003

3 GHz Pentium 43 GHz Pentium 4

Windows XPWindows XP

– Files (1 B, 1 MB, 3 MB, 68 MB)Files (1 B, 1 MB, 3 MB, 68 MB)– Algorithms and keysAlgorithms and keys

DES: 64 bitsDES: 64 bits

3DES: 128 and 192 bits3DES: 128 and 192 bits

AES: 128, 192, and 256 bitsAES: 128, 192, and 256 bits

RSA: 512 and 1024 bitsRSA: 512 and 1024 bits

Page 8: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

88

Performance MeasurementsPerformance MeasurementsRSA vs. Other Algorithms

Using Polynomial Fitted Lines (n=2)3 GHz Pentium 4

0

20

40

60

80

100

120

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5

File Size (106 Bytes)

Tim

e (

s)

DES, 3DES and AES

RSA - 512 bit Encryption

RSA - 512 bit Decryption

RSA - 1024 bit Encryption

RSA - 1024 bit Decryption

Page 9: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

99

Performance MeasurementsPerformance MeasurementsEncryption and Decryption AveragesUsing Polynomial Fitted Lines (n=2)

3 GHz Pentium 4

0

2

4

6

8

10

12

14

0 10 20 30 40 50 60 70

File Size (106 Bytes)

Tim

e (

s)

DES - 56 bit3DES - 112 bit3DES - 168 bitAES - 128 bitAES - 192 bitAES - 256 bit

Page 10: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1010

Performance MeasurementsPerformance Measurements

Throughputs – 3 GHzThroughputs – 3 GHz– SymmetricSymmetric

Percent ofAlgorithm MB/s Fastest AlgorithmDES 64-bit 8.10 100.00%AES 128-bit 7.08 87.40%3DES 128-bit 6.90 85.15%3DES 192-bit 6.80 84.01%AES 192-bit 6.52 80.54%AES 256-bit 6.10 75.28%

– Public KeyPublic Key

Percent of Percent ofEncryption MB/s Fastest Algorithm Decryption MB/s Fastest AlgorithmRSA 512-bit 0.90 11.11% RSA 512-bit 0.11 1.36%RSA 1024-bit 0.62 10.17% RSA 1024-bit 0.04 0.49%

Page 11: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1111

Performance MeasurementsPerformance Measurements

AnalysisAnalysis– Curious how much of the performance was Curious how much of the performance was

due to the encryption vs. how much was due due to the encryption vs. how much was due to system overhead (e.g., file system)to system overhead (e.g., file system)

Repeated study on a slower machineRepeated study on a slower machine– 600 MHz Pentium 3600 MHz Pentium 3– Windows XPWindows XP

Page 12: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1212

Performance MeasurementsPerformance MeasurementsEncryption and Decryption AveragesUsing Polynomial Fitted Lines (n=2)

600 MHz Pentium 3

0

10

20

30

40

50

60

0 10 20 30 40 50 60 70

File Size (106 Bytes)

Tim

e (

s)

DES - 56 bit3DES - 112 bit3DES - 168 bitAES - 128 bitAES - 192 bitAES - 256 bit

Page 13: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1313

Performance MeasurementsPerformance Measurements

Throughputs – 600 MHzThroughputs – 600 MHz– SymmetricSymmetric

Percent ofAlgorithm MB/s Fastest AlgorithmDES 64-bit 2.45 100.00%AES 128-bit 1.71 69.90%3DES 192-bit 1.67 68.21%3DES 128-bit 1.67 68.04%AES 192-bit 1.52 62.06%AES 256-bit 1.39 56.74%

– Public KeyPublic Key

Percent of Percent ofEncryption MB/s Fastest Algorithm Decryption MB/s Fastest AlgorithmRSA 512-bit 0.28 11.42% RSA 512-bit 0.03 1.22%RSA 1024-bit 0.21 8.57% RSA 1024-bit 0.01 0.41%

Page 14: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1414

RecommendationsRecommendations

Use managed code (C#)Use managed code (C#)Use AES with 256-bit keysUse AES with 256-bit keys

RationaleRationale– code safetycode safety– modularity of encryption servicemodularity of encryption service– suitability as a web service in .NETsuitability as a web service in .NET– AES-256 performance was within 20% of DESAES-256 performance was within 20% of DES– exponentially more secure than any other algorithmexponentially more secure than any other algorithm– protection against the unknown (e.g., progress in protection against the unknown (e.g., progress in

quantum computing)quantum computing)

Page 15: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1515

Workflow ModelWorkflow Model

Department of Radiology ModelDepartment of Radiology Model

Data

Data

Reports

DICOMWorklist

HL7HL7

PatientsHospital

RegistrationSystem

ExamScheduleSystem

ImageModality

RISHIS

DICOMGateway

ReportingSystem

Workstation

PACSArchive

RelationalDatabase

HL7 HL7

HL7HL7

HL7Reports

DICOM

DICOM

Page 16: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1616

Workflow ModelWorkflow Model

Involved StepsInvolved Steps

Steps Description A Patient Registration by hospital registration system B Notify HIS of patient and data using Health Level 7 (HL7) C Schedule exam and notify RIS D Patient data to RIS and to PACS archive E DICOM worklist to image modality F Conduct patient exam G Patient image data to gateway using DICOM H Relational data to gateway (required prior images) I DICOM image data from gateway to PACS archive J DICOM image data to workstation from PACS archive K Patient report generated in reporting system L Patient report send to RIS from reporting system M Patient report sent from RIS to HIS

Page 17: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1717

Workflow ModelWorkflow Model

ResourcesResources

Resource Description R1 Hospital Registration System R2 HIS (Hospital Information System) R3 RIS (Radiology Information System) R4 Examination Schedule System R5 HL7 Communications for Text Data R6 DICOM Communications for Image Data R7 Image Modality Unit R8 DICOM Gateway R9 Relational Database R10 PACS Archive R11 Workstation R12 Reporting System R13 Encryption/Decryption Application

Page 18: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1818

Workflow ModelWorkflow Model

Bottleneck Table – From Resource Allocation TableBottleneck Table – From Resource Allocation Table

Bottleneck Equation B1 1 / (T1 + T2) B2 1 / (T2 + T4 + T13) B3 1 / (T3 + T4 + T5 + T12 + T13) B4 1 / (T3) B5 1 / (T2 + T3 + T4 + T12 + T13) B6 1 / (T5 + T7 + T8 + T9 + T10) B7 1 / (T5 + T6 + T7) B8 1 / (T7 + T8 + T9) B9 1 / (T8) B10 1 / (T4 + T9 + T10) B11 1 / (T10) B12 1 / (T11 + T12) B13 1 / (T4 + T7 + T8 + T9 + T10)

Page 19: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

1919

Workflow ModelWorkflow Model

Bottleneck CalculationBottleneck Calculation– was Bwas B77, the Image Modality Unit, the Image Modality Unit

Bottleneck Equation Without Encryption With Encryption B1 1 / (T1 + T2) 3.98 3.98 B2 1 / (T2 + T4 + T13) 79.92 78.26 B3 1 / (T3 + T4 + T5 + T12 + T13) 32.73 32.43 B4 1 / (T3) 120.00 120.00 B5 1 / (T2 + T3 + T4 + T12 + T13) 34.29 33.96 B6 1 / (T5 + T7 + T8 + T9 + T10) 5.37 3.96 B7 1 / (T5 + T6 + T7) 2.59 2.48 B8 1 / (T7 + T8 + T9) 6.67 5.00 B9 1 / (T8) 20.00 15.00 B10 1 / (T4 + T9 + T10) 11.61 8.35 B11 1 / (T10) 30.00 20.00 B12 1 / (T11 + T12) 24.00 24.00 B13 1 / (T4 + T7 + T8 + T9 + T10) N/A 3.95

Throughput Throughput Patients/HrPatients/Hr

Page 20: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

2020

Workflow ModelWorkflow Model

Throughput ResultsThroughput Results– Sequential Patient ModelSequential Patient Model

7% Performance Degradation7% Performance Degradation

– Highly Concurrent Patient ModelHighly Concurrent Patient Model5% Performance Degradation5% Performance Degradation

Reassuring to determine that HIPAA's impact is Reassuring to determine that HIPAA's impact is modestmodestPossible to recover throughput through other Possible to recover throughput through other optimizations in patient flowoptimizations in patient flow

Page 21: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

2121

Workflow ModelWorkflow Model

BoundsBounds– Infinite ResourcesInfinite Resources

N / (TN / (Tee + T + Tss))

– Bottleneck LimitBottleneck Limit1 / T1 / Tbb

– Upper BoundUpper BoundN / (TN / (Tee + T + Tss + (N – 1) * T + (N – 1) * Tbb))

– Lower BoundLower Bound1 / (T1 / (Tee + T + Tss))

TTee = Time Spent Encrypting = Time Spent Encrypting

TTss = Total System Time – T = Total System Time – Tee

TTbb = Time Spent on = Time Spent on

Bottleneck StepBottleneck Step

Page 22: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

2222

Workflow ModelWorkflow ModelSystem with EncryptionSystem with Encryption

Page 23: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

2323

SummarySummaryImpact of HIPAA's encryption requirements were Impact of HIPAA's encryption requirements were initially unknowninitially unknown

Suitability of web services approach untestedSuitability of web services approach untested

Public key algorithm (RSA) unsuitablePublic key algorithm (RSA) unsuitable

Three symmetric key algorithms (DES, 3DES, AES) Three symmetric key algorithms (DES, 3DES, AES) were all suitablewere all suitable

AES-256 encrypts a 500-slice MR file of 68 MB in 12 AES-256 encrypts a 500-slice MR file of 68 MB in 12 seconds on a 3 GHz Pentium 4seconds on a 3 GHz Pentium 4

Workflow model using AES-256 predicts a patient Workflow model using AES-256 predicts a patient throughput reduction of 5-7%throughput reduction of 5-7%

Now have an understanding of workflow and where Now have an understanding of workflow and where to optimizeto optimize

Page 24: 1 The e-Logistics of Securing Distributed Medical Data Andrew M. Snyder Alfred C. Weaver

2424

AcknowledgementsAcknowledgements

Funding for this work is provided by:Funding for this work is provided by:

David Ladd and Tom HealyUniversity Research Program

Microsoft ResearchMicrosoft Corporation