Presentation by:
PRACHI PANCHOLI
Assistant Professor | Computer Engineering
L.D. COLLEGE OF ENGINEERING , AHMEDABAD
Human Skin Modeling using Deep
Learning
ID:PHD21376
RESEARCH PROPOSAL
1
Background Problem
Introduction
State of the Art Hierarchy
Literature Survey
Current Scope
Research Objectives
References
PHD21376
02
01
03
04
05
06
07
08
Content
09
10
Research Scope
Proposed Plan
2
Future Scope
11 Questions
Introduction (Why Skin is chosen as a topic area)
• Skin-the largest organ of the human body
• A Barrier- its function to protect the human body against pathogens
• Skin diseases
 the fourth most common cause of all human disease
 affecting almost one-third of the world’s population- major causes of the global disease burden [GDB]
 covers all cultural regions and occurs in all ages
 Impacts social, financial and psychological consequences on the patient’s life and society[2]
• Skin cancer is ranked second for causes of worldwide death (except COVID-19 in 2020)
• Skin lesion detection follows :
 ABCDE rule( asymmetry, border, color, dermoscopic structure and evolving)
 CASH rule (color ,architecture , symmetry and homogeneity)
3
PHD21376
4
Situation
Challenge
Background Problem
5
Solution
Background Problem(Contd..)
State –of-art hierarchy
[
6
PHD21376
Literature Survey
7
PHD21376
Paper Model /Classifier + Data Sets
used
Features examined Limitations identified
“Classification of skin lesions
using transfer learning and
augmentation with Alex-
net” [7]
DCNN AlexNet , over 2500+
images were taken from DERM IS
,MED-NODE and ISIC Challege
data set 2017
AlexNet model is built and
different parameters are set.
Replacement of the
classification layer with a
softmax layer that works with
two or three kinds of skin
lesions, and augmenting
dataset by fixed and random
rotation angles
“Automatic diagnosis of skin
diseases using convolution
neural network”[6]
CNN AlexNet , 174 images were
taken from DERM Net
The Convolutional Neural
Network (CNN) used in this
paper has utilized around 11
layers viz., Convolution Layer,
Activation Layer, Pooling Layer,
Fully Connected Layer and
Soft-Max Classifier.
The proposed methodogy does
not apply to more class of
diseases , also other challenges
automating the process
includes the variation of skin
tones, location of the disease,
specifications of the image
acquisition system etc
Literature Survey(Contd..)
8
PHD21376
Paper Model /Classifier + Data Sets
used
Features examined Limitations identified
“Man against machine:
diagnostic performance
of a deep learning
convolutional neural
network for
dermoscopic melanoma
recognition in
comparison to 58
dermatologists”[5]
CNN Inception v4 , 100-image
test-set was used (level-I:
dermoscopy only; level-II:
dermoscopy plus clinical
information and images
Proposed model was trained and validated
using dermoscopic images and
corresponding diagnoses. the results were
compared with 58 dermatologists
Test-sets used did not display the
full range of lesions and poor
availability of validated images led
to a shortage of melanocytic
lesions from other skin types and
genetic backgrounds.
“Modeling of Human
Skin by the Use of Deep
Learning”[4]
ISVM , 15000 images were
taken from DERMNZ
Early detection of skin tumors is done by
proposing computer view detection system
for segmentation and recognizing skin
diseases by extracting set of features from
given skin lesions image to produce
appropriate classification
Extraction of any new feature and
training on different dataset is not
done. Also multiple skin lesions
cannot be identified .
“Discriminative Feature
Learning for Skin
Disease
Classification Using
Deep Convolutional
Neural Network”[8]
CNN ResNet 152 and Inception
ResNet-V2, 12000+ faciual skin
disease images were
considered (further labelling is
done by Wuhan Union
Hospital)
Triplet Loss function is consider to interact
capability of performance management
with an extreme of labels also The proposed
method performs layer-wise fine-tuning of
pre-train deep CNN models, instead of block
wise , to improve the performance of the
end-to-end learning method.
Only facial images were analysed .
A properly designeddataset with
the help of a dermatologist and
organized taxonomy is required.
Literature Survey(Contd..)
9
PHD21376
Paper Model /Classifier +
Data Sets used
Features examined Limitations identified
“Classification of Skin
Disease Using Deep
Learning Neural
Networks with MobileNet
V2 and LSTM”[10]
DNN MobielNet V2 +
LSTM, 10000 images
were considered
HAM10000(Kaggle)
dataset.
The MobileNet V2 architecture is designed to
work with a portable device with a stride2
mechanism. the use of the LSTM module with the
MobileNet V2 would enhance the prediction
accuracy by maintaining the previous timestamp
data.
The proposed approach is not
designed to replace but rather to
supplement existing disease-
diagnostic solutions. Laboratory test
results are always more
trustworthy than diagnoses based
solely on visual symptoms, and visual
inspection alone often challenges
early diagnosis
“A convolutional neural
network trained with
dermoscopic images
performed on par with
145 dermatologists in a
clinical melanoma image
classification Task”[9]
CNN ResNet , 12,378
open-source
dermoscopic images
This study compares the performance of a
convolutional neuronal network trained with
dermoscopic images exclusively for identifying
melanoma in clinical photographs with the
manual grading of the same images by
dermatologists
The training, validation and test set
were derived from a mostly fair
skinned population, and the data sets
were comprised of images typically
seen in clinical practice . The proposed
model would perform better if the
clinical data is improvised
“Deep Learning in Skin
Disease Image
Recognition: A Review”[1]
CNN AlexNet,
VGG, Inception,
ResNet, and
DensenNet
A survey has been done for 45 papers and
analysis from the aspects of disease type, data
set,data processing technology, data
augmentation technology, model for skin disease
image recognition, deep learning framework,
evaluation indicators, and model performance is
done
The interpretability research on skin
disease identification models is a big
gap in this survey ; also Embedding AI
diagnosis and treatment of skin
disease on smart devices should be
considered.
Current Scope
Where (This research is currently being used and which area deals with it)
• Currently, Machine Learning models are being used for Disease classification using clinical and derma
pathology images [11]
• Deep Learning plays vital role for accurate feature extraction of lesions [1]
1
0
PHD21376
Specific Field
Used Dataset
Limited to Dermatology
• Research work + Limited
accuracy of Diagnoses
Quality
Parameters
Feature Extraction (Color + Texture)
• Accuracy + Sensitivity +
Specificity
• Derm Quest
• DermNZ
• DermIS
Research Scope and goals
Scope:
 Assist dermatologist in classifying multiple skin lesions ( especially with more classifier category )
Goal
 Use state -of-art technique, called Deep learning to design an intelligent and adaptive medical-image
based skin lesion diagnosis system
 Achieve (or improve upon) results for:
 Skin lesions segmentation and
 Skin lesion classification
 Evaluate the impact of multiple skin lesions segmentation on the accuracy of the classifier
1
1
PHD21376
Research Objectives
Examine the medical image in dermatology:
• The primary stage is to improve the image classification performance
 Two-stage image synthesis(Augmentation) and enhancement approach(Super resolution)
 Applying image classification problems with traditional deep learning algorithms.
• WSOD(XAI) will be used to disclose the performance of the black-box deep
learning model.
• In the Secondary Stage, an object detection approach will be used with anchor
boxes and compared with WSOD.
12
PHD21376
Future Scope
How (this work will have impact on future scenarios)
• Interpretability study on skin disease identification [3]
• Embedding AI diagnosis and treatment of skin disease on smart devices will be a significant
trend in the future. [1]
• Clinical identification of skin disease [1]
13
PHD21376
Proposed Plan of work
What (are the things or activities needed to be done in this research )
14
PHD21376
Work Timeline
15
PHD21376
References
[1] Ling-fang Li1 , Xu Wang1 , Wei-jian Hu 1 , Neal N. XIONG 2 , (Senior Member, IEEE), YONG-XING DU
1 , AND BAO-SHAN LI1, ” Deep Learning in Skin Disease Image Recognition: A Review”, IEEE Access -
November 11, 2020, date of current version December 1, 2020., Volume-8
[2] Nazia Hameed 1,* , Fozia Hameed 2 , Antesar Shabut 3 , Sehresh Khan 4 , Silvia Cirstea 5 and
Alamgir Hossain 6,”An Intelligent Computer-Aided Scheme for Classifying Multiple Skin Lesions”,
Computers 2019, 8, 62; doi:10.3390/computers8030062
[3] Anjali Gautam, Balasubramanian Raman,“Towards accurate classification of skin cancer from
dermatology images”, IET Image Processing, DOI: 10.1049/ipr2.12166, Accepted: 5 February 2021
[4] Xin Xiong,1 Xuexun Guo,2 and Yiping Wang , “Modeling of Human Skin by the Use of Deep Learning”,
Hindawi Complexity Volume 2021, Article ID 5531585, 11 pages https://doi.org/10.1155/2021/5531585
16
PHD21376
References
[5] H. A. Haenssle1, C. Fink, R. Schneiderbauer, F.Toberer,T. Buhl , A. Blum , A. Kalloo , A. Ben Hadj
Hassen , L. Thomas , A. Enk & L. Uhlmann, “Man against machine: diagnostic performance of a deep
learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58
dermatologists”, Annals of Oncology 29: 1836–1842, 2018 doi:10.1093/annonc/mdy166 Published
online 28 May 2018
[6] T. Shanthi a,R.S. Sabeenianb , R. Anand,“Automatic diagnosis of skin diseases using convolution
neural network”, https://doi.org/10.1016/j.micpro.2020.103074 0141-9331/© 2020 Elsevier,
Microprocessors and Microsystems 76 (2020) 103074
[7] Khalid M. HosnyID1 *, Mohamed A. KassemID2 , Mohamed M. Foaud,“Classification of skin
lesions using transfer learning and augmentation with Alex-net”, PLOS ONE |
https://doi.org/10.1371/journal.pone.0217293 May 21, 2019
[8] BELAL AHMAD 1 , MOHD USAMA1 , CHUEN-MIN HUANG2 , KAI HWANG 3 , (Life Fellow, IEEE),
M. SHAMIM HOSSAIN 4 , (Senior Member, IEEE), AND GHULAM MUHAMMAD, “Discriminative
Feature Learning for Skin Disease Classification Using Deep Convolutional Neural Network”,IEEE
Access, Digital Object Identifier 10.1109/ACCESS.2020.2975198
17
PHD21376
References
[9] Titus J. Brinker a,b, *, Achim Hekler a , Alexander H. Enk b, “A convolutional neural network trained
with dermoscopic images performed on par with 145 dermatologists in a clinical melanoma image
classification task”, Europian Journal of Cancer 111(2019) 148-154, Science Direct,
https://doi.org/10.1016/j.ejca.2019.02.005 0959-8049/ª 2019
[10] Parvathaneni Naga Srinivasu, alluri Gnana SivaSai , Muhammad Fazal Ijaz 3, Akash Kumar Bhoi
, Wonjoon Kim and James Jin Kang,” Classification of Skin Disease Using Deep Learning Neural
Networks with MobileNet V2 and LSTM”, Sensors 2021, 21, 2852. https://doi.org/10.3390/s21082852
[11] Stephanie Chan . Vidhatha Reddy . Bridget Myers . Quinn Thibodeaux . Nicholas Brownstone .
Wilson Liao, ” Machine Learning in Dermatology: Current Applications, Opportunities, and
Limitations”, Dermatol Ther (Heidelb) (2020) 10:365–386 https://doi.org/10.1007/s13555-020-00372-0
18
PHD21376
QuestionsYou
Soil Sense
Growing Prosperity
(Feel Free To Shoot Questions)
19

SKin lesion detection using ml approach.pptx

  • 1.
    Presentation by: PRACHI PANCHOLI AssistantProfessor | Computer Engineering L.D. COLLEGE OF ENGINEERING , AHMEDABAD Human Skin Modeling using Deep Learning ID:PHD21376 RESEARCH PROPOSAL 1
  • 2.
    Background Problem Introduction State ofthe Art Hierarchy Literature Survey Current Scope Research Objectives References PHD21376 02 01 03 04 05 06 07 08 Content 09 10 Research Scope Proposed Plan 2 Future Scope 11 Questions
  • 3.
    Introduction (Why Skinis chosen as a topic area) • Skin-the largest organ of the human body • A Barrier- its function to protect the human body against pathogens • Skin diseases  the fourth most common cause of all human disease  affecting almost one-third of the world’s population- major causes of the global disease burden [GDB]  covers all cultural regions and occurs in all ages  Impacts social, financial and psychological consequences on the patient’s life and society[2] • Skin cancer is ranked second for causes of worldwide death (except COVID-19 in 2020) • Skin lesion detection follows :  ABCDE rule( asymmetry, border, color, dermoscopic structure and evolving)  CASH rule (color ,architecture , symmetry and homogeneity) 3 PHD21376
  • 4.
  • 5.
  • 6.
  • 7.
    Literature Survey 7 PHD21376 Paper Model/Classifier + Data Sets used Features examined Limitations identified “Classification of skin lesions using transfer learning and augmentation with Alex- net” [7] DCNN AlexNet , over 2500+ images were taken from DERM IS ,MED-NODE and ISIC Challege data set 2017 AlexNet model is built and different parameters are set. Replacement of the classification layer with a softmax layer that works with two or three kinds of skin lesions, and augmenting dataset by fixed and random rotation angles “Automatic diagnosis of skin diseases using convolution neural network”[6] CNN AlexNet , 174 images were taken from DERM Net The Convolutional Neural Network (CNN) used in this paper has utilized around 11 layers viz., Convolution Layer, Activation Layer, Pooling Layer, Fully Connected Layer and Soft-Max Classifier. The proposed methodogy does not apply to more class of diseases , also other challenges automating the process includes the variation of skin tones, location of the disease, specifications of the image acquisition system etc
  • 8.
    Literature Survey(Contd..) 8 PHD21376 Paper Model/Classifier + Data Sets used Features examined Limitations identified “Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists”[5] CNN Inception v4 , 100-image test-set was used (level-I: dermoscopy only; level-II: dermoscopy plus clinical information and images Proposed model was trained and validated using dermoscopic images and corresponding diagnoses. the results were compared with 58 dermatologists Test-sets used did not display the full range of lesions and poor availability of validated images led to a shortage of melanocytic lesions from other skin types and genetic backgrounds. “Modeling of Human Skin by the Use of Deep Learning”[4] ISVM , 15000 images were taken from DERMNZ Early detection of skin tumors is done by proposing computer view detection system for segmentation and recognizing skin diseases by extracting set of features from given skin lesions image to produce appropriate classification Extraction of any new feature and training on different dataset is not done. Also multiple skin lesions cannot be identified . “Discriminative Feature Learning for Skin Disease Classification Using Deep Convolutional Neural Network”[8] CNN ResNet 152 and Inception ResNet-V2, 12000+ faciual skin disease images were considered (further labelling is done by Wuhan Union Hospital) Triplet Loss function is consider to interact capability of performance management with an extreme of labels also The proposed method performs layer-wise fine-tuning of pre-train deep CNN models, instead of block wise , to improve the performance of the end-to-end learning method. Only facial images were analysed . A properly designeddataset with the help of a dermatologist and organized taxonomy is required.
  • 9.
    Literature Survey(Contd..) 9 PHD21376 Paper Model/Classifier + Data Sets used Features examined Limitations identified “Classification of Skin Disease Using Deep Learning Neural Networks with MobileNet V2 and LSTM”[10] DNN MobielNet V2 + LSTM, 10000 images were considered HAM10000(Kaggle) dataset. The MobileNet V2 architecture is designed to work with a portable device with a stride2 mechanism. the use of the LSTM module with the MobileNet V2 would enhance the prediction accuracy by maintaining the previous timestamp data. The proposed approach is not designed to replace but rather to supplement existing disease- diagnostic solutions. Laboratory test results are always more trustworthy than diagnoses based solely on visual symptoms, and visual inspection alone often challenges early diagnosis “A convolutional neural network trained with dermoscopic images performed on par with 145 dermatologists in a clinical melanoma image classification Task”[9] CNN ResNet , 12,378 open-source dermoscopic images This study compares the performance of a convolutional neuronal network trained with dermoscopic images exclusively for identifying melanoma in clinical photographs with the manual grading of the same images by dermatologists The training, validation and test set were derived from a mostly fair skinned population, and the data sets were comprised of images typically seen in clinical practice . The proposed model would perform better if the clinical data is improvised “Deep Learning in Skin Disease Image Recognition: A Review”[1] CNN AlexNet, VGG, Inception, ResNet, and DensenNet A survey has been done for 45 papers and analysis from the aspects of disease type, data set,data processing technology, data augmentation technology, model for skin disease image recognition, deep learning framework, evaluation indicators, and model performance is done The interpretability research on skin disease identification models is a big gap in this survey ; also Embedding AI diagnosis and treatment of skin disease on smart devices should be considered.
  • 10.
    Current Scope Where (Thisresearch is currently being used and which area deals with it) • Currently, Machine Learning models are being used for Disease classification using clinical and derma pathology images [11] • Deep Learning plays vital role for accurate feature extraction of lesions [1] 1 0 PHD21376 Specific Field Used Dataset Limited to Dermatology • Research work + Limited accuracy of Diagnoses Quality Parameters Feature Extraction (Color + Texture) • Accuracy + Sensitivity + Specificity • Derm Quest • DermNZ • DermIS
  • 11.
    Research Scope andgoals Scope:  Assist dermatologist in classifying multiple skin lesions ( especially with more classifier category ) Goal  Use state -of-art technique, called Deep learning to design an intelligent and adaptive medical-image based skin lesion diagnosis system  Achieve (or improve upon) results for:  Skin lesions segmentation and  Skin lesion classification  Evaluate the impact of multiple skin lesions segmentation on the accuracy of the classifier 1 1 PHD21376
  • 12.
    Research Objectives Examine themedical image in dermatology: • The primary stage is to improve the image classification performance  Two-stage image synthesis(Augmentation) and enhancement approach(Super resolution)  Applying image classification problems with traditional deep learning algorithms. • WSOD(XAI) will be used to disclose the performance of the black-box deep learning model. • In the Secondary Stage, an object detection approach will be used with anchor boxes and compared with WSOD. 12 PHD21376
  • 13.
    Future Scope How (thiswork will have impact on future scenarios) • Interpretability study on skin disease identification [3] • Embedding AI diagnosis and treatment of skin disease on smart devices will be a significant trend in the future. [1] • Clinical identification of skin disease [1] 13 PHD21376
  • 14.
    Proposed Plan ofwork What (are the things or activities needed to be done in this research ) 14 PHD21376
  • 15.
  • 16.
    References [1] Ling-fang Li1, Xu Wang1 , Wei-jian Hu 1 , Neal N. XIONG 2 , (Senior Member, IEEE), YONG-XING DU 1 , AND BAO-SHAN LI1, ” Deep Learning in Skin Disease Image Recognition: A Review”, IEEE Access - November 11, 2020, date of current version December 1, 2020., Volume-8 [2] Nazia Hameed 1,* , Fozia Hameed 2 , Antesar Shabut 3 , Sehresh Khan 4 , Silvia Cirstea 5 and Alamgir Hossain 6,”An Intelligent Computer-Aided Scheme for Classifying Multiple Skin Lesions”, Computers 2019, 8, 62; doi:10.3390/computers8030062 [3] Anjali Gautam, Balasubramanian Raman,“Towards accurate classification of skin cancer from dermatology images”, IET Image Processing, DOI: 10.1049/ipr2.12166, Accepted: 5 February 2021 [4] Xin Xiong,1 Xuexun Guo,2 and Yiping Wang , “Modeling of Human Skin by the Use of Deep Learning”, Hindawi Complexity Volume 2021, Article ID 5531585, 11 pages https://doi.org/10.1155/2021/5531585 16 PHD21376
  • 17.
    References [5] H. A.Haenssle1, C. Fink, R. Schneiderbauer, F.Toberer,T. Buhl , A. Blum , A. Kalloo , A. Ben Hadj Hassen , L. Thomas , A. Enk & L. Uhlmann, “Man against machine: diagnostic performance of a deep learning convolutional neural network for dermoscopic melanoma recognition in comparison to 58 dermatologists”, Annals of Oncology 29: 1836–1842, 2018 doi:10.1093/annonc/mdy166 Published online 28 May 2018 [6] T. Shanthi a,R.S. Sabeenianb , R. Anand,“Automatic diagnosis of skin diseases using convolution neural network”, https://doi.org/10.1016/j.micpro.2020.103074 0141-9331/© 2020 Elsevier, Microprocessors and Microsystems 76 (2020) 103074 [7] Khalid M. HosnyID1 *, Mohamed A. KassemID2 , Mohamed M. Foaud,“Classification of skin lesions using transfer learning and augmentation with Alex-net”, PLOS ONE | https://doi.org/10.1371/journal.pone.0217293 May 21, 2019 [8] BELAL AHMAD 1 , MOHD USAMA1 , CHUEN-MIN HUANG2 , KAI HWANG 3 , (Life Fellow, IEEE), M. SHAMIM HOSSAIN 4 , (Senior Member, IEEE), AND GHULAM MUHAMMAD, “Discriminative Feature Learning for Skin Disease Classification Using Deep Convolutional Neural Network”,IEEE Access, Digital Object Identifier 10.1109/ACCESS.2020.2975198 17 PHD21376
  • 18.
    References [9] Titus J.Brinker a,b, *, Achim Hekler a , Alexander H. Enk b, “A convolutional neural network trained with dermoscopic images performed on par with 145 dermatologists in a clinical melanoma image classification task”, Europian Journal of Cancer 111(2019) 148-154, Science Direct, https://doi.org/10.1016/j.ejca.2019.02.005 0959-8049/ª 2019 [10] Parvathaneni Naga Srinivasu, alluri Gnana SivaSai , Muhammad Fazal Ijaz 3, Akash Kumar Bhoi , Wonjoon Kim and James Jin Kang,” Classification of Skin Disease Using Deep Learning Neural Networks with MobileNet V2 and LSTM”, Sensors 2021, 21, 2852. https://doi.org/10.3390/s21082852 [11] Stephanie Chan . Vidhatha Reddy . Bridget Myers . Quinn Thibodeaux . Nicholas Brownstone . Wilson Liao, ” Machine Learning in Dermatology: Current Applications, Opportunities, and Limitations”, Dermatol Ther (Heidelb) (2020) 10:365–386 https://doi.org/10.1007/s13555-020-00372-0 18 PHD21376
  • 19.