SlideShare a Scribd company logo
1 of 27
Classification of Tuberculosis Type on CT Scans of
Lungs using a fusion of 2D and 3D Deep
Convolutional Neural Networks
Emad Aghajanzadeh, Behzad Shomali, Diba Aminshahidi, Navid Ghassemi
CLEF2021 Working Notes
09/22/2021
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
2
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
3
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Our team
Emad Aghajanzadeh Behzad Shomali Diba Aminshahidi Navid Ghassemi
Computer Engineering
Undergraduate Student
Computer Engineering
Undergraduate Student
Computer Engineering
Alumna
Computer Engineering
Graduate Student
4
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
5
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Ranking
The best run ranked:
• 4th based on the Kappa (value of 0.181)
• 3rd based on the Accuracy (value of 0.404)
6
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
7
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Task / data
• Classification of CT scans based on their TB type
• 5 existing classes
• 917 training, 421 test images
8
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
9
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Preprocessing
• HU values are scaled to be in range [0, 1]
• Keep 50 middle slices
• Resize slices to 100x100
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 10
Preprocessing
There were two challenges while training:
• Small dataset
• Data duplication / augmentation
• Rotate by a degree in range [-5, 5]
• Zoom by a ratio of 1.25
• Resize to original size
• Imbalanced dataset
• Variable penalties
• Remove some samples
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 11
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 12
Proposed methods (settings)
• Adam Optimizer (default parameters)
• Train/validation split ratio of 0.2
• Batch size:
• 32 for 2D CNNs
• 8 for 3D CNNs
• Loss function
• Combination of Cross-entropy and weighted Kappa
• Change the multipliers over time
13
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (2D CNN)
14
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (2D CNN)
• Assign each CT label to all its slices
• Feed each slice to 2D CNN separately
• A vector of 50 labels for each CT
• Pick the most appearing label
• Pick the label with highest probability
15
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (3D CNN)
16
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (3D CNN)
• Input: 100x100x50
• Capture the spatial information
17
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (2D CNN + RNN)
18
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
• Used the best-trained 2D CNN
• Feeding the:
• Predicted labels by the 2D CNN to RNN
• Features extracted by the last layer of the 2D CNN to RNN
• The results were almost the same to 2D CNN
Proposed methods (2D CNN + RNN)
19
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (Fusion of 2D & 3D CNN with RNN)
20
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (Fusion of 2D & 3D CNN with RNN)
• 3D CNN can capture the spatial information
• 2D CNN can better extract features in each slice
• Concatenate the features extracted by each model
• Feed the feature vector to RNN
• The result was similar to that of the 3D CNN
• Implies the more contribution of the 3D model
21
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (Fusion of 2D & 3D + manual)
By investigating the confusion matrix, we found out:
• 3D models can better classify the last 3 classes
• 2D models is good at classifying the first 2 classes
• Select the final prediction manually
• Have a list of predicted labels for each CT scan
• Select based on the models with lower variance
22
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
Proposed methods (results)
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 23
Evaluation result on 2D CNN
Evaluation result on 3D CNN
Final evaluation result on test data
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 24
Conclusion & Future works
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 25
• Proposed different approaches
• 2D CNN didn’t perform well
• Apply a smarter voting mechanism
• Define a fixed size window which moves through the labels vector
• Use more complicated data augmentation approaches
• Train a separate binary classification
Contents
• Our team
• Ranking
• Task / data
• Preprocessing
• Proposed methods
• Conclusion & Future works
• Questions
Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 26
Thanks for your
consideration
{emad.aghajanzadeh, behzad.shomali, d.aminshahidi, navidghassemi}@mail.um.ac.ir

More Related Content

Similar to Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks

Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
sugiuralab
 
Unknown power power point unknown power point
Unknown power power point unknown power pointUnknown power power point unknown power point
Unknown power power point unknown power point
xmendquick
 

Similar to Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks (20)

Current developments in video quality: From the emerging HEVC standard to tem...
Current developments in video quality: From the emerging HEVC standard to tem...Current developments in video quality: From the emerging HEVC standard to tem...
Current developments in video quality: From the emerging HEVC standard to tem...
 
Project Dr Deeper
Project Dr DeeperProject Dr Deeper
Project Dr Deeper
 
An optimized deep learning architecture for the diagnosis of covid 19 disease...
An optimized deep learning architecture for the diagnosis of covid 19 disease...An optimized deep learning architecture for the diagnosis of covid 19 disease...
An optimized deep learning architecture for the diagnosis of covid 19 disease...
 
Deep Conditional Adversarial learning for polyp Segmentation
Deep Conditional Adversarial learning for polyp SegmentationDeep Conditional Adversarial learning for polyp Segmentation
Deep Conditional Adversarial learning for polyp Segmentation
 
Performance Comparison Analysis for Medical Images Using Deep Learning Approa...
Performance Comparison Analysis for Medical Images Using Deep Learning Approa...Performance Comparison Analysis for Medical Images Using Deep Learning Approa...
Performance Comparison Analysis for Medical Images Using Deep Learning Approa...
 
diagnostics-11-02208-v2.pdf
diagnostics-11-02208-v2.pdfdiagnostics-11-02208-v2.pdf
diagnostics-11-02208-v2.pdf
 
Cycle-free CycleGAN using invertible generator for unsupervised low-dose CT d...
Cycle-free CycleGAN using invertible generator for unsupervised low-dose CT d...Cycle-free CycleGAN using invertible generator for unsupervised low-dose CT d...
Cycle-free CycleGAN using invertible generator for unsupervised low-dose CT d...
 
Springsim 2016 data driven healthcare tutorial
Springsim 2016 data driven healthcare tutorialSpringsim 2016 data driven healthcare tutorial
Springsim 2016 data driven healthcare tutorial
 
AAPM REPORT 186
AAPM REPORT 186AAPM REPORT 186
AAPM REPORT 186
 
Dual energy CT in radiotherapy: Current applications and future outlook
Dual energy CT in radiotherapy: Current applications and future outlookDual energy CT in radiotherapy: Current applications and future outlook
Dual energy CT in radiotherapy: Current applications and future outlook
 
Covid-19 Detection Using CNN MODEL
Covid-19 Detection Using CNN MODELCovid-19 Detection Using CNN MODEL
Covid-19 Detection Using CNN MODEL
 
Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
Cervical Spine Range of Motion Measurement Utilizing Image Analysis - VISAPP2022
 
CovidAID: COVID-19 Detection using Chest X-Ray Images
CovidAID: COVID-19 Detection using Chest X-Ray ImagesCovidAID: COVID-19 Detection using Chest X-Ray Images
CovidAID: COVID-19 Detection using Chest X-Ray Images
 
Machine Learning approaches at video compression
Machine Learning approaches at video compression Machine Learning approaches at video compression
Machine Learning approaches at video compression
 
IRJET- A Survey on Medical Image Interpretation for Predicting Pneumonia
IRJET- A Survey on Medical Image Interpretation for Predicting PneumoniaIRJET- A Survey on Medical Image Interpretation for Predicting Pneumonia
IRJET- A Survey on Medical Image Interpretation for Predicting Pneumonia
 
Unknown power power point unknown power point
Unknown power power point unknown power pointUnknown power power point unknown power point
Unknown power power point unknown power point
 
Cancerous lung nodule detection in computed tomography images
Cancerous lung nodule detection in computed tomography imagesCancerous lung nodule detection in computed tomography images
Cancerous lung nodule detection in computed tomography images
 
Deep Learning_Blog.pdf
Deep Learning_Blog.pdfDeep Learning_Blog.pdf
Deep Learning_Blog.pdf
 
Final_Project_ESE_441
Final_Project_ESE_441Final_Project_ESE_441
Final_Project_ESE_441
 
Computer-aided diagnostic system kinds and pulmonary nodule detection efficacy
Computer-aided diagnostic system kinds and pulmonary nodule  detection efficacyComputer-aided diagnostic system kinds and pulmonary nodule  detection efficacy
Computer-aided diagnostic system kinds and pulmonary nodule detection efficacy
 

Recently uploaded

If this Giant Must Walk: A Manifesto for a New Nigeria
If this Giant Must Walk: A Manifesto for a New NigeriaIf this Giant Must Walk: A Manifesto for a New Nigeria
If this Giant Must Walk: A Manifesto for a New Nigeria
Kayode Fayemi
 
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
Sheetaleventcompany
 

Recently uploaded (20)

BDSM⚡Call Girls in Sector 97 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 97 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 97 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 97 Noida Escorts >༒8448380779 Escort Service
 
The workplace ecosystem of the future 24.4.2024 Fabritius_share ii.pdf
The workplace ecosystem of the future 24.4.2024 Fabritius_share ii.pdfThe workplace ecosystem of the future 24.4.2024 Fabritius_share ii.pdf
The workplace ecosystem of the future 24.4.2024 Fabritius_share ii.pdf
 
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
Governance and Nation-Building in Nigeria: Some Reflections on Options for Po...
 
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 93 Noida Escorts >༒8448380779 Escort Service
 
Microsoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AIMicrosoft Copilot AI for Everyone - created by AI
Microsoft Copilot AI for Everyone - created by AI
 
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptxMohammad_Alnahdi_Oral_Presentation_Assignment.pptx
Mohammad_Alnahdi_Oral_Presentation_Assignment.pptx
 
SaaStr Workshop Wednesday w/ Lucas Price, Yardstick
SaaStr Workshop Wednesday w/ Lucas Price, YardstickSaaStr Workshop Wednesday w/ Lucas Price, Yardstick
SaaStr Workshop Wednesday w/ Lucas Price, Yardstick
 
If this Giant Must Walk: A Manifesto for a New Nigeria
If this Giant Must Walk: A Manifesto for a New NigeriaIf this Giant Must Walk: A Manifesto for a New Nigeria
If this Giant Must Walk: A Manifesto for a New Nigeria
 
George Lever - eCommerce Day Chile 2024
George Lever -  eCommerce Day Chile 2024George Lever -  eCommerce Day Chile 2024
George Lever - eCommerce Day Chile 2024
 
Introduction to Prompt Engineering (Focusing on ChatGPT)
Introduction to Prompt Engineering (Focusing on ChatGPT)Introduction to Prompt Engineering (Focusing on ChatGPT)
Introduction to Prompt Engineering (Focusing on ChatGPT)
 
Mathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptxMathematics of Finance Presentation.pptx
Mathematics of Finance Presentation.pptx
 
VVIP Call Girls Nalasopara : 9892124323, Call Girls in Nalasopara Services
VVIP Call Girls Nalasopara : 9892124323, Call Girls in Nalasopara ServicesVVIP Call Girls Nalasopara : 9892124323, Call Girls in Nalasopara Services
VVIP Call Girls Nalasopara : 9892124323, Call Girls in Nalasopara Services
 
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
No Advance 8868886958 Chandigarh Call Girls , Indian Call Girls For Full Nigh...
 
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night EnjoyCall Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
Call Girl Number in Khar Mumbai📲 9892124323 💞 Full Night Enjoy
 
Air breathing and respiratory adaptations in diver animals
Air breathing and respiratory adaptations in diver animalsAir breathing and respiratory adaptations in diver animals
Air breathing and respiratory adaptations in diver animals
 
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
Re-membering the Bard: Revisiting The Compleat Wrks of Wllm Shkspr (Abridged)...
 
ANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docxANCHORING SCRIPT FOR A CULTURAL EVENT.docx
ANCHORING SCRIPT FOR A CULTURAL EVENT.docx
 
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
Andrés Ramírez Gossler, Facundo Schinnea - eCommerce Day Chile 2024
 
Presentation on Engagement in Book Clubs
Presentation on Engagement in Book ClubsPresentation on Engagement in Book Clubs
Presentation on Engagement in Book Clubs
 
Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510Thirunelveli call girls Tamil escorts 7877702510
Thirunelveli call girls Tamil escorts 7877702510
 

Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks

  • 1. Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks Emad Aghajanzadeh, Behzad Shomali, Diba Aminshahidi, Navid Ghassemi CLEF2021 Working Notes 09/22/2021
  • 2. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions 2 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 3. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions 3 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 4. Our team Emad Aghajanzadeh Behzad Shomali Diba Aminshahidi Navid Ghassemi Computer Engineering Undergraduate Student Computer Engineering Undergraduate Student Computer Engineering Alumna Computer Engineering Graduate Student 4 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 5. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions 5 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 6. Ranking The best run ranked: • 4th based on the Kappa (value of 0.181) • 3rd based on the Accuracy (value of 0.404) 6 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 7. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions 7 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 8. Task / data • Classification of CT scans based on their TB type • 5 existing classes • 917 training, 421 test images 8 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 9. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions 9 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 10. Preprocessing • HU values are scaled to be in range [0, 1] • Keep 50 middle slices • Resize slices to 100x100 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 10
  • 11. Preprocessing There were two challenges while training: • Small dataset • Data duplication / augmentation • Rotate by a degree in range [-5, 5] • Zoom by a ratio of 1.25 • Resize to original size • Imbalanced dataset • Variable penalties • Remove some samples Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 11
  • 12. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 12
  • 13. Proposed methods (settings) • Adam Optimizer (default parameters) • Train/validation split ratio of 0.2 • Batch size: • 32 for 2D CNNs • 8 for 3D CNNs • Loss function • Combination of Cross-entropy and weighted Kappa • Change the multipliers over time 13 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 14. Proposed methods (2D CNN) 14 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 15. Proposed methods (2D CNN) • Assign each CT label to all its slices • Feed each slice to 2D CNN separately • A vector of 50 labels for each CT • Pick the most appearing label • Pick the label with highest probability 15 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 16. Proposed methods (3D CNN) 16 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 17. Proposed methods (3D CNN) • Input: 100x100x50 • Capture the spatial information 17 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 18. Proposed methods (2D CNN + RNN) 18 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 19. • Used the best-trained 2D CNN • Feeding the: • Predicted labels by the 2D CNN to RNN • Features extracted by the last layer of the 2D CNN to RNN • The results were almost the same to 2D CNN Proposed methods (2D CNN + RNN) 19 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 20. Proposed methods (Fusion of 2D & 3D CNN with RNN) 20 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 21. Proposed methods (Fusion of 2D & 3D CNN with RNN) • 3D CNN can capture the spatial information • 2D CNN can better extract features in each slice • Concatenate the features extracted by each model • Feed the feature vector to RNN • The result was similar to that of the 3D CNN • Implies the more contribution of the 3D model 21 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 22. Proposed methods (Fusion of 2D & 3D + manual) By investigating the confusion matrix, we found out: • 3D models can better classify the last 3 classes • 2D models is good at classifying the first 2 classes • Select the final prediction manually • Have a list of predicted labels for each CT scan • Select based on the models with lower variance 22 Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes
  • 23. Proposed methods (results) Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 23 Evaluation result on 2D CNN Evaluation result on 3D CNN Final evaluation result on test data
  • 24. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 24
  • 25. Conclusion & Future works Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 25 • Proposed different approaches • 2D CNN didn’t perform well • Apply a smarter voting mechanism • Define a fixed size window which moves through the labels vector • Use more complicated data augmentation approaches • Train a separate binary classification
  • 26. Contents • Our team • Ranking • Task / data • Preprocessing • Proposed methods • Conclusion & Future works • Questions Classification of Tuberculosis Type on CT Scans of Lungs using a fusion of 2D and 3D Deep Convolutional Neural Networks CLEF2021 Working Notes 26
  • 27. Thanks for your consideration {emad.aghajanzadeh, behzad.shomali, d.aminshahidi, navidghassemi}@mail.um.ac.ir