Lecture Subject
Internet of Things (IoT) and AI Role in Medical
and Healthcare Systems
Somayyeh Jafarali Jassbi
PhD, Assistant Professor,
Head of Computer Engineering Department, SRBIAU
s.jassbi@srbiau.ac.ir
Hamidreza Bolhasani
PhD Candidate, Fonder & Data Scientist at DataBioX
hamidreza.bolhasani@srbiau.ac.ir
March 2022
- Internet of Things (IoT) History
- IoT Applications for Healthcare
- Internet of Nano Things (IoNT)
- IoNT Applications in Health and Medical
- IoT Systems Architecture Overview
- IoT and Artificial Intelligence (AI)
- AI / ML / DL Fundamental Concepts
- Deep Learning Accelerators
- Conclusion - Q & A
Table of Contents
Internet of Things (IoT)
1
Ref. Mahmuda Khatun Mishu et al. MDPI 2020. Prospective Ecient Ambient Energy Harvesting
Sources for IoT-Equipped Sensor Applications
IoT for Healthcare
2
Ref.
[1] Yazdan Ahmad Qadri et al. IEEE COMST 2020
[2] Kyeonghye Guk et al. MDPI, Nanomaterial, 2019.
IoT for Healthcare
3
Ref.
Hadi Habibzadeh et al. IEEE Internet of Things Journal, 2019
IoT for Healthcare
4
IoT for Healthcare
5
6
Internet of Nano Things (IoNT)
Ref.
Kyeonghye Guk et al. MDPI, Nanomaterial, 2019.
7
Internet of Nano Things (IoNT)
8
Internet of Nano Things (IoNT)
9
Internet of Nano Things (IoNT)
10
IoT and Artificial Intelligence (AI)
Ref.
Hadi Habibzadeh et al. IEEE Internet of Things Journal, 2019
11
IoT and Artificial Intelligence (AI)
Ref.
Yazdan Ahmad Qadri et al. IEEE COMST 2020
Artificial Intelligence (AI)
12
Deep Neural Networks (DNN)
Ref. B. Reagen, R. Adolf, P. N. Whatmough, G. Wei, and D. M. Brooks, Deep Learning for Computer Architects,
ser. Synthesis Lectures on Computer Architecture. Morgan & Claypool Publishers, 2017.
13
Biological Neuron
Ref. Jennifer Walinga, 2014. 14
Ref. Jennifer Walinga, 2014.
Biological Neuron
15
Biological Neuron
Ref. Prof. Dr. Jurgen Brauer, “Introduction to Deep Learning,” 2018. 16
Artificial Neural Networks (ANN)
Ref. Prof. Dr. Jurgen Brauer, “Introduction to Deep Learning,” 2018. 17
Artificial Neural Networks (ANN)
18
Artificial Neural Networks (ANN)
19
Deep Neural Networks
20
Deep Neural Networks
21
22
Deep Learning: Anywhere!
Ref. Dr. Dara Rahmati, Hardware Accelerator Course Slides, 2020. 23
24
25
Convolutional Neural Networks (CNNs)
Ref. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in
International Conference on Learning Representations (ICLR), 2015. 26
Convolutional Neural Networks (CNNs)
Ref. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in
International Conference on Learning Representations (ICLR), 2015.
AlexNet
VGG16
27
Histopathologic Images CT-Scan MRI PET-Scan
What does a Deep Neural Network needs to be Trained?
BIG DATA
How much data?
So much that we get our desired results!
28
Our First Contribution
DataBioX founded in 2020 for sharing biomedical datasets.
DataBioX first dataset: Invasive Ductal Carcinoma Grading Dataset
29
19
Grade I Grade II Grade III
Invasive Ductal Carcinoma Grading Dataset
DataBioX: Training dataset for Supervised Learning
31
Invasive Ductal Carcinoma Grading Dataset
DataBioX: Training dataset for Supervised Learning
32
Published Paper in 2020
Elsevier Publication
Officially requested by many prestigious
Universities around the world like MIT and
University of Johns Hopkins.
33
Breast Cancer Classification Dataset
BreaKHis: Training dataset for Supervised Learning
cmccool@uni-bonn.de
34
Breast Cancer Detection
Deep Learning Applied for Histological Diagnosis of Breast Cancer (2020)
35
Breast Cancer Detection
Deep Learning to Improve Breast Cancer Detection on Screening Mammography (2019)
36
Breast Cancer Treatment Plan
Early prediction of neoadjuvant chemotherapy response for advanced breast cancer
Using PET/MRI image deep learning
37
Colon Cancer Analysis
Deep learning for colon cancer histopathological images analysis (2021)
38
Glioblastoma Detection
Analyzing MRI scans to detect glioblastoma tumor using hybrid deep belief networks (2020)
39
Ref. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in
International Conference on Learning Representations (ICLR), 2015.
40
Fundamental Concepts
Convolutional Neural Networks (CNNs) Computing Architecture
Why is this important?
The big opportunity: Data Reuse
Ref. Michael Pellauer, et al., “MODELING AND ANALYZING DEEP
LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018.
41
Fundamental Concepts
Data Movement and its costs in CNN Architectures
- Fetch Data
- Multiply and Accumulation
Data Flow
+
Computing
Localization
Data Reuse
Ref. Michael Pellauer, et al., “MODELING AND ANALYZING DEEP
LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018.
42
Fundamental Concepts
Data Flows, a way for Easy and cheap access to the data
Spatial Architecture
Data Flow
Temporal Architecture
SIMD/SIMT
43
Fundamental Concepts
Data Flows | Spatial and Temporal Architetcures
Ref. Michael Pellauer, et al., “MODELING AND ANALYZING DEEP
LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018.
44
Fundamental Concepts
Data Flows | a 7D Computational Problem
Input-Stationary (IS) Data Flow
Ref. Joel Emer, et al., “Tutorial on Hardware Accelerators for Deep Neural Networks,” ISCA Tutorial, June 2019.
45
Output-Stationary (OS) Data Flow
Weight-Stationary (WS) Data Flow
Row-Stationary (RS) Data Flow
46
Fundamental Concepts
NVDLA Deep Learning Accelerator | WS Data Flow
47
Fundamental Concepts
Eyeriss Deep Learning Accelerator | RS Data Flow
Input Parameter
No.
LayerFile
1
dataFlowFile
2
vectorWidth
3
NoCBandwidth
4
multicastSupported
5
numAverageHopsinNoC
6
numPEs
7
48
Simulation Tools
MAESTRO
27
Buffer Analysis
Data Flow NLR NVDLA
L1 Buffer Requirement
(Byte)
18.00 66.00
L2 Buffer Requirement
(KB)
1.12 4.12
L1RdSum 7,225,344 451,584
L1WrSum 7,225,344 451,584
L2RdSum 462,422,016 28,901,376
L2WrSum 462,422,016 28,901,376
L1 Weight Reuse 1 16
L1 Input Reuse 4 16
L2 Weight Reuse 448 190.26
L2 Input Reuse 2,633 4,473
NoC Analysis
L1 to L2 NoC BW 128 32
L2 to L1 NoC BW 160 1,024
Performance Analysis
L1 to L2 Sum 56 32
L1 to L2 Delay 4.43 4.25
L2 to L1 Delay 0 0
Roofline Throughput
(GFLOPS with 1 GHZ
clock)
896 128
Compute Runtime 169 421
Total Runtime (Cycles) 1,428,553,728 384,072,192
1,428,553,728
384,072,192
0
200,000,000
400,000,000
600,000,000
800,000,000
1,000,000,000
1,200,000,000
1,400,000,000
1,600,000,000
Total runtime (Cycles)
NLR vgg16_conv11 NVDLA vgg16_conv11
1
4
16 16
0
2
4
6
8
10
12
14
16
18
L1 weight reuse L1 input reuse
NLR vgg16_conv11 NVDLA vgg16_conv11
49
DLA Performance Evaluation
Thanks
s.jassbi@srbiau.ac.ir
hamidreza.bolhasani@srbiau.ac.ir

Internet of Things (IoT) and Artificial Intelligence (AI) role in Medical and Healthcare Systems

  • 1.
    Lecture Subject Internet ofThings (IoT) and AI Role in Medical and Healthcare Systems Somayyeh Jafarali Jassbi PhD, Assistant Professor, Head of Computer Engineering Department, SRBIAU s.jassbi@srbiau.ac.ir Hamidreza Bolhasani PhD Candidate, Fonder & Data Scientist at DataBioX hamidreza.bolhasani@srbiau.ac.ir March 2022
  • 2.
    - Internet ofThings (IoT) History - IoT Applications for Healthcare - Internet of Nano Things (IoNT) - IoNT Applications in Health and Medical - IoT Systems Architecture Overview - IoT and Artificial Intelligence (AI) - AI / ML / DL Fundamental Concepts - Deep Learning Accelerators - Conclusion - Q & A Table of Contents
  • 3.
    Internet of Things(IoT) 1 Ref. Mahmuda Khatun Mishu et al. MDPI 2020. Prospective Ecient Ambient Energy Harvesting Sources for IoT-Equipped Sensor Applications
  • 4.
    IoT for Healthcare 2 Ref. [1]Yazdan Ahmad Qadri et al. IEEE COMST 2020 [2] Kyeonghye Guk et al. MDPI, Nanomaterial, 2019.
  • 5.
    IoT for Healthcare 3 Ref. HadiHabibzadeh et al. IEEE Internet of Things Journal, 2019
  • 6.
  • 7.
  • 8.
    6 Internet of NanoThings (IoNT) Ref. Kyeonghye Guk et al. MDPI, Nanomaterial, 2019.
  • 9.
    7 Internet of NanoThings (IoNT)
  • 10.
    8 Internet of NanoThings (IoNT)
  • 11.
    9 Internet of NanoThings (IoNT)
  • 12.
    10 IoT and ArtificialIntelligence (AI) Ref. Hadi Habibzadeh et al. IEEE Internet of Things Journal, 2019
  • 13.
    11 IoT and ArtificialIntelligence (AI) Ref. Yazdan Ahmad Qadri et al. IEEE COMST 2020
  • 14.
  • 15.
    Deep Neural Networks(DNN) Ref. B. Reagen, R. Adolf, P. N. Whatmough, G. Wei, and D. M. Brooks, Deep Learning for Computer Architects, ser. Synthesis Lectures on Computer Architecture. Morgan & Claypool Publishers, 2017. 13
  • 16.
  • 17.
    Ref. Jennifer Walinga,2014. Biological Neuron 15
  • 18.
    Biological Neuron Ref. Prof.Dr. Jurgen Brauer, “Introduction to Deep Learning,” 2018. 16
  • 19.
    Artificial Neural Networks(ANN) Ref. Prof. Dr. Jurgen Brauer, “Introduction to Deep Learning,” 2018. 17
  • 20.
  • 21.
  • 22.
  • 23.
  • 24.
  • 25.
    Deep Learning: Anywhere! Ref.Dr. Dara Rahmati, Hardware Accelerator Course Slides, 2020. 23
  • 26.
  • 27.
  • 28.
    Convolutional Neural Networks(CNNs) Ref. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in International Conference on Learning Representations (ICLR), 2015. 26
  • 29.
    Convolutional Neural Networks(CNNs) Ref. K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in International Conference on Learning Representations (ICLR), 2015. AlexNet VGG16 27
  • 30.
    Histopathologic Images CT-ScanMRI PET-Scan What does a Deep Neural Network needs to be Trained? BIG DATA How much data? So much that we get our desired results! 28
  • 31.
    Our First Contribution DataBioXfounded in 2020 for sharing biomedical datasets. DataBioX first dataset: Invasive Ductal Carcinoma Grading Dataset 29
  • 32.
  • 33.
    Grade I GradeII Grade III Invasive Ductal Carcinoma Grading Dataset DataBioX: Training dataset for Supervised Learning 31
  • 34.
    Invasive Ductal CarcinomaGrading Dataset DataBioX: Training dataset for Supervised Learning 32
  • 35.
    Published Paper in2020 Elsevier Publication Officially requested by many prestigious Universities around the world like MIT and University of Johns Hopkins. 33
  • 36.
    Breast Cancer ClassificationDataset BreaKHis: Training dataset for Supervised Learning cmccool@uni-bonn.de 34
  • 37.
    Breast Cancer Detection DeepLearning Applied for Histological Diagnosis of Breast Cancer (2020) 35
  • 38.
    Breast Cancer Detection DeepLearning to Improve Breast Cancer Detection on Screening Mammography (2019) 36
  • 39.
    Breast Cancer TreatmentPlan Early prediction of neoadjuvant chemotherapy response for advanced breast cancer Using PET/MRI image deep learning 37
  • 40.
    Colon Cancer Analysis Deeplearning for colon cancer histopathological images analysis (2021) 38
  • 41.
    Glioblastoma Detection Analyzing MRIscans to detect glioblastoma tumor using hybrid deep belief networks (2020) 39
  • 42.
    Ref. K. Simonyanand A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” in International Conference on Learning Representations (ICLR), 2015. 40 Fundamental Concepts Convolutional Neural Networks (CNNs) Computing Architecture
  • 43.
    Why is thisimportant? The big opportunity: Data Reuse Ref. Michael Pellauer, et al., “MODELING AND ANALYZING DEEP LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018. 41 Fundamental Concepts Data Movement and its costs in CNN Architectures
  • 44.
    - Fetch Data -Multiply and Accumulation Data Flow + Computing Localization Data Reuse Ref. Michael Pellauer, et al., “MODELING AND ANALYZING DEEP LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018. 42 Fundamental Concepts Data Flows, a way for Easy and cheap access to the data
  • 45.
    Spatial Architecture Data Flow TemporalArchitecture SIMD/SIMT 43 Fundamental Concepts Data Flows | Spatial and Temporal Architetcures
  • 46.
    Ref. Michael Pellauer,et al., “MODELING AND ANALYZING DEEP LEARNING ACCELERATOR DATAFLOWS WITH MAESTRO,” in MAERI Tutorial, 2018. 44 Fundamental Concepts Data Flows | a 7D Computational Problem
  • 47.
    Input-Stationary (IS) DataFlow Ref. Joel Emer, et al., “Tutorial on Hardware Accelerators for Deep Neural Networks,” ISCA Tutorial, June 2019. 45 Output-Stationary (OS) Data Flow Weight-Stationary (WS) Data Flow Row-Stationary (RS) Data Flow
  • 48.
    46 Fundamental Concepts NVDLA DeepLearning Accelerator | WS Data Flow
  • 49.
    47 Fundamental Concepts Eyeriss DeepLearning Accelerator | RS Data Flow
  • 50.
  • 51.
    27 Buffer Analysis Data FlowNLR NVDLA L1 Buffer Requirement (Byte) 18.00 66.00 L2 Buffer Requirement (KB) 1.12 4.12 L1RdSum 7,225,344 451,584 L1WrSum 7,225,344 451,584 L2RdSum 462,422,016 28,901,376 L2WrSum 462,422,016 28,901,376 L1 Weight Reuse 1 16 L1 Input Reuse 4 16 L2 Weight Reuse 448 190.26 L2 Input Reuse 2,633 4,473 NoC Analysis L1 to L2 NoC BW 128 32 L2 to L1 NoC BW 160 1,024 Performance Analysis L1 to L2 Sum 56 32 L1 to L2 Delay 4.43 4.25 L2 to L1 Delay 0 0 Roofline Throughput (GFLOPS with 1 GHZ clock) 896 128 Compute Runtime 169 421 Total Runtime (Cycles) 1,428,553,728 384,072,192 1,428,553,728 384,072,192 0 200,000,000 400,000,000 600,000,000 800,000,000 1,000,000,000 1,200,000,000 1,400,000,000 1,600,000,000 Total runtime (Cycles) NLR vgg16_conv11 NVDLA vgg16_conv11 1 4 16 16 0 2 4 6 8 10 12 14 16 18 L1 weight reuse L1 input reuse NLR vgg16_conv11 NVDLA vgg16_conv11 49 DLA Performance Evaluation
  • 52.