Reference 1

# An Interpretable Skin Cancer Classification Using

Optimized Convolutional Neural Network for a Smart
Healthcare System
Abstract:
Skin cancer is a prevalent form of malignancy globally, and its early and accurate diagnosis is
critical for patient survival. Clinical evaluation of skin lesions is essential, but it faces
challenges such as long waiting times and subjective interpretations. Deep learning techniques have
been developed to tackle these challenges and assist dermatologists in making more accurate
diagnoses. Prompt treatment of skin cancer is vital to prevent its progression and potentially
life-threatening consequences. The use of deep learning algorithms can improve the speed and
accuracy of diagnosis, leading to earlier detection and treatment. Additionally, it can reduce the
workload for healthcare professionals, allowing them to concentrate on more complex cases. The goal
of this study was to develop reliable deep learning (DL) prediction models for skin cancer
classification; (i) deal with a typical severe class imbalance problem, which arises because the
skin-affected patients’ class is significantly smaller than the healthy class; and (ii) interpret
the model output to better understand the decision-making mechanism (iii) Propose an End-to-End
smart healthcare system through an android application. In a comparison examination with six well-
known classifiers, the effectiveness of the proposed DL technique was explored in terms of metrics
relating to both generalization capability and classification accuracy. A study used the HAM10000
dataset and an optimized CNN to identify the seven forms of skin cancer. The model was trained
using two optimization functions (Adam and RMSprop) and three activation functions (Relu, Swish,
and Tanh). Furthermore, an XAI-based skin lesion classification system was developed, incorporating
Grad-CAM and Grad-CAM++ to explain the model’s decisions. This system can help doctors make
informed skin cancer diagnoses in their early stages, with an 82% classification accuracy and 0.47%
loss accuracy.

Conclusion
A clinical decision support system’s skin image classifier can serve as a second opinion for
dermatologists. Although a large research community has helped, these AI-based systems can only
make predictions and cannot explain their rationale. This is where XAI approaches come in. We
demonstrated how to approach skin image analysis in a domain-specific manner. For example, if a
dermatologist identifies a lesion as a nevus but the model labels it as a melanoma, both the doctor
and the patient may wonder “Why?” Our method includes explanations such as “if the color of the
lesion is constant, the classifier’s confidence in melanoma diagnosis drops.” The clinician may
then notice the color irregularity in the dermatoscopic picture, which is not evident on the
lesion, and figure out why the classifier predicted incorrectly. Whether the clinical decision
support system supports or opposes the physician’s diagnosis, offering human-readable reasons
fosters confidence and improves system knowledge. Furthermore, our perturbation-based explanation
technique for diagnosis employing medically relevant and irrelevant characteristics may have
implications in other medical domains. In Table 14, we are showing the implications of our model to
other medical images dataset. The dataset is just tested by our model. In the future, we will train
our model by using attention/transformer so that the interpretability score get increase.

Skin Cancer Detection Using Combined Decision of Deep

Learners
Abstract:
Cancer is a deadly disease that arises due to the growth of uncontrollable body cells. Every year,
a large number of people succumb to cancer and it’s been labeled as the most serious public health
snag. Cancer can develop in any part of the human anatomy, which may consist of trillions of
cellules. One of the most frequent type of cancer is skin cancer which develops in the upper layer
of the skin. Previously, machine learning techniques have been used for skin cancer detection using
protein sequences and different kinds of imaging modalities. The drawback of the machine learning
approaches is that they require human-engineered features, which is a very laborious and time-
taking activity. Deep learning addressed this issue to some extent by providing the facility of
automatic feature extraction. In this study, convolution-based deep neural networks have been used
for skin cancer detection using ISIC public dataset. Cancer detection is a sensitive issue, which
is prone to errors if not timely and accurately detected. The performance of the individual machine
learning models to detect cancer is limited. The combined decision of individual learners is
expected to be more accurate than the individual learners. The ensemble learning technique exploits
the diversity of learners to yield a better decision. Thus, the prediction accuracy can be enhanced
by combing the decision of individual learners for sensitive issues such as cancer detection. In
this paper, an ensemble of deep learners has been developed using learners of VGG, CapsNet, and
ResNet for skin cancer detection. The results show that the combined decision of deep learners is
superior to the finding of individual learners in terms of sensitivity, accuracy, specificity, F-
score, and precision. The experimental results of this study provide a compelling reason to be
applied for other disease detection.

Conclusion
The malignant lesion is the leading cause of death due to skin cancer. If it is diagnosed in the
early stages its treatment may be possible. In literature, deep learning approaches have been used
to detect cancer but the performance of the individual learners is limited. The performance can be
enhanced by combining the decision of diverse individual learners for decision-making on sensitive
issues such as cancer. This paper developed an ensemble model to detect skin cancer. It is
developed by combining decision the three deep learning models of VGG, Caps-Net, and ResNet. It is
noticed from the results that the proposed ensemble achieved an average accuracy of 93.5% with a
classification training time of 106 s. The proposed model performs better than individual learners
with respect to different quality measures i.e. sensitivity, accuracy, F-Score, specificity, false-
Positive, and precision. In future, we are intend to study the achievement of reinforcement
learning-based techniques for skin cancer detection.

The Power of Generative AI to Augment for Enhanced Skin

Cancer Classification: A Deep Learning Approach
Abstract:
Skin cancer, particularly the malignant melanoma subtype, is widely recognized as a highly lethal
form of cancer characterized by abnormal melanocyte cell growth. However, diagnosing and
classifying skin lesions, as well as automatically recognizing malignant tumors from dermoscopy
images, present significant challenges. To address this challenge, our study employs variants of
Convolutional Neural Networks (CNNs) to effectively diagnose and classify various skin lesion types
using the latest benchmark datasets ISIC 2019 and 2020. The dataset underwent rigorous
preprocessing, which involves employing advanced Generative Artificial Intelligence (AI) techniques
i.e., Generative Adversarial Networks (GANs) and Enhanced Super-Resolution Generative Adversarial
Networks (ESRGAN), for augmentation. These generative techniques are carefully evaluated and
compared for their effectiveness. Our CNN-based approach involves aggregating results from multiple
transfer learning models, including VGG16, VGG19, SVM along with a hybrid model in combination of
VGG19 and SVM. On ISIC 2019, we have achieved promising accuracies of 92% for VGG16 and 93% for
VGG19. Notably, the hybrid VGG19+SVM model exhibits the highest accuracy of 96%. On ISIC 2020,
VGG16, VGG19, and SVM achieves accuracies of 90%, 92%, and 92%, respectively. Our findings
underscore the potential of generative AI for augmentation, and the efficacy of CNN-based transfer
learning models in improving skin cancer classification accuracy.

Conclusion
Skin cancer, recognized as one of the most deadly forms of cancer, poses a significant threat to
individuals worldwide. Among its types, malignant melanoma stands out as a particularly dreadful
form, characterized by the abnormal growth of melanocyte cells. While skin lesions are prevalent,
accurately characterizing them and automating the identification of malignant tumors from
dermoscopy images remain complex challenges in the field.
This study aimed to address these challenges by leveraging convolutional neural networks (CNNs) to
diagnose and classify various types of skin lesions, using the extensive ISIC 2019 and ISIC 2020
datasets. To prepare the data for analysis, a comprehensive preprocessing pipeline is implemented,
including image augmentation, normalization, and resizing. Notably, advanced techniques such as
ImageDataGenerator and GAN are employed for effective data augmentation, and the resulting outcomes
are thoroughly compared. The proposed CNN-based approach involved aggregating results from multiple
iterations, enabling the enhancement of overall classification accuracy. Moreover, an extensive
evaluation of different transfer learning models, including VGG16, VGG19, SVM, and a hybrid
VGG19+SVM model, is conducted to assess their efficacy in skin lesion classification.
The experimental results on the ISIC 2019 dataset showcased promising accuracy levels, with the
VGG16 and VGG19 models achieving accuracies of 92% and 93% respectively. Particularly noteworthy is
the hybrid VGG19+SVM model, which exhibited the highest accuracy of 96%. On the ISIC 2020 dataset,
the VGG16, VGG19 and SVM models achieved accuracies of 90%, 92% and 92% respectively. These
findings underscore the potential of CNN-based approaches in the accurate diagnosis and
classification of skin lesions. Leveraging transfer learning models, such as VGG16, VGG19, SVM, and
the hybrid VGG19+SVM model, proved instrumental in achieving higher classification accuracy. This
study contributes valuable insights to the field of automated skin cancer detection and
classification, paving the way for improved diagnosis and treatment strategies.

ECRNet: Hybrid Network for Skin Cancer Identification

Abstract:
Skin cancer recognition poses a significant challenge in the field of deep learning. While
conventional convolutional neural networks have been extensively employed for classifying skin
cancer images, their fixed receptive field limits their ability to capture the global features
present in such images. Conversely, transformer-based models that rely on self-attention can
effectively model long-range dependencies, but they come with high computational complexity and
exhibit certain limitations in local feature induction. To address this issue, this paper presents
a novel skin cancer recognition network named ECRNet. ECRNet has been designed to effectively
capture both global and local information, and it introduces an explicit vision center to
accomplish this purpose. Moreover, this paper presents a feature fusion module known as the CCPA
block. This module utilizes both coordinate attention and channel attention mechanisms to extract
image features and enhance the representation of skin cancer images. To evaluate the performance of
ECRNet, extensive experimental comparisons were conducted on the ISIC2018 dataset. The experimental
results demonstrate that ECRNet outperforms the baseline model, showing improvements of 1.19% in
accuracy (ACC), 1.96% in precision, 4.08% in recall, and 3.28% in the F1 score.

Conclusion
In this study, a novel attention module called CCPA (Coordinated Channel and Position Attention) is
proposed. The CCPA module combines coordinate attention and channel attention while incorporating
partial convolution. This module effectively addresses the issue of insufficient inter-channel
information and improves the spatial feature extraction capability. Additionally, the CCPA module
is integrated with the EVC (Enhanced Visual Context) module and residual module to construct a new
skin cancer recognition model named ECRNet. This integration further enhances the capability of
skin cancer image recognition.
Experimental results demonstrate that the ECRNet model achieves outstanding performance in skin
cancer image recognition tasks, surpassing traditional CNN and ViT (Vision Transformer) methods.
However, our model has not reached the desired depth, which directly limits its performance in
terms of accuracy and precision. In order to improve the overall effectiveness of the model, we
must conduct more in-depth research and exploration on the depth and accuracy of the model. In
addition, we plan to further study Convnext-tiny in combination with CCPA and EVC modules to
improve the recognition of high-noise images and achieve more significant improvements in future
work.
Abstract:
Skin cancer is a senior public health issue that could profit from computer-aided diagnosis to
decrease the encumbrance of this widespread disease. Researchers have been more motivated to
develop computer-aided diagnosis systems because visual examination wastes time. The initial stage
in skin lesion analysis is skin lesion segmentation, which might assist in the following
categorization task. It is a difficult task because sometimes the whole lesion might be the same
colors, and the borders of pigment regions can be foggy. Several studies have effectively handled
skin lesion segmentation; nevertheless, developing new methodologies to improve efficiency is
necessary. This work thoroughly analyzes the most advanced algorithms and methods for skin lesion
segmentation. The review begins with traditional segmentation techniques, followed by a brief
review of skin lesion segmentation using deep learning and optimization techniques. The main
objective of this work is to highlight the strengths and weaknesses of a wide range of algorithms.
Additionally, it examines various commonly used datasets for skin lesions and the metrics used to
evaluate the performance of these techniques.

Conclusion and Future Directions

Melanoma, a dangerous type of disease, is causing increased deaths yearly. In recent years,
melanoma has become one of the most common reasons for human death. In the case of early detection,
patients will have a better chance of surviving. So, the accuracy of computerized melanoma
detection becomes more and more important. Detection of melanoma begins with skin image pre-
processing, followed by segmentation. The skin lesion classification may be erroneous if the lesion
segmentation is not carried out appropriately.
The chance of detecting melanoma is decreased if the segmentation is performed poorly. Therefore,
this paper presents an analytical survey of the major pigmented skin lesion segmentation
techniques. The literature survey analysis shows that researchers developed and applied various
techniques. These techniques cover pre-processing and segmentation techniques of skin lesion
images. Deep learning, Optimization-based, and Optimized Deep learning methods were examined.
The literature survey analysis clearly shows that the researchers developed and applied various
computational approaches. However, among them, the rising of deep learning-based and optimized deep
learning image segmentation techniques is noticeable since several public datasets have ground
truth images. Deep learning-based and optimized techniques are frequently employed for lesion
segmentation, producing highly promising segmented outcomes. The important advantage of using the
optimization techniques is that it reduces the time complexity and helps increase efficiency
without degrading the quality of the image. Nature-inspired optimization algorithms have been used
for multilevel thresholding or clustering skin lesion segmentation and are effective and achieve
high results compared to traditional algorithms. It is also noticed that traditional techniques
(edge-, region-, thresholding-based) approaches are also used but not significantly in this domain.
Deep learning has common issues, including network structure design, 3D data image segmentation
model, and loss function design. Designing 3D convolution models to analyze 3D skin lesion image
data is a researchable direction. Loss function design has long been a challenge in deep learning
research. Optimized Deep Learning models solve these problems.
In addition to segmentation techniques, this research looked at the dataset(s) that the authors
utilized in their publication and when training their models. Based on the accuracy achieved by
their segmentation technique, we also did a comparison analysis of the utilized research
publications. It has also been noticed that the high usage of the PH2 dataset. In addition to PH2,
the ISIC 2016 and 2017 datasets have been utilized significantly. However, ISIC 2019 and 2020
datasets should be widely used in the future.
The following are some significant future directions:
Enhancing image quality with advanced techniques can also improve performance, in addition to
the development of CNN models. It will also be possible to segment lesions using an embedding
system automatically.
Different combinations of layers and classifiers can be explored to improve the accuracy of the
image segmentation model. An efficient solution is still required to improve the image
 segmentation model’s performance. So, the various new deep learning model designs can be
explored by future researchers.
Mobile dermoscopic image analysis: With various inexpensive dermoscopic designed for
smartphones, mobile dermoscopic image analysis is of great interest worldwide, especially in
regions with limited access to dermatologists. Typical DL-based image segmentation algorithms
have millions of weights. In addition, classical CNN architectures are known to show difficulty
in dealing with certain image errors, such as noise and blur. Furthermore, it has been shown
that DL-based skin lesion diagnostic models are vulnerable to similar artifacts: different kinds
of noise and blur, brightness and contrast changes, dark corners, bubbles, rulers, ink markings,
etc. Therefore, the current dermoscopic image segmentation algorithms may not be ideal for
execution on typically resource-constrained mobile and edge devices needed for patient privacy
so that uploading skin images to remote servers is avoided.

DeepSkin: A Deep Learning Approach for Skin Cancer

Classification
Abstract:
Skin cancer is one of the most rapidly spreading illnesses in the world and because of the limited
resources available. Early detection of skin cancer is crucial accurate diagnosis of skin cancer
identification for preventive approach in general. Detecting skin cancer at an early stage is
challenging for dermatologists, as well in recent years, both supervised and unsupervised learning
tasks have made extensive use of deep learning. One of these models, Convolutional Neural Networks
(CNN), has surpassed all others in object detection and classification tests. The dataset is
screened from MNIST: HAM10000 which consists of seven different types of skin lesions with the
sample size of 10015 is used for the experimentation. The data pre-processing techniques like
sampling, dull razor and segmentation using autoencoder and decoder is employed. Transfer learning
techniques like DenseNet169 and Resnet 50 were used to train the model to obtain the results.

Conclusion
Skin cancer is one of the illnesses that is spreading the quickest on the earth. Skin cancer is
mostly brought on by a person’s vulnerability to the sun’s UV radiation. Given the limited
resources available, early identification of skin cancer is essential. Accurate diagnosis and
identification viability are generally essential for skin cancer prevention strategies.
Additionally, dermatologists have trouble seeing skin cancer in its early stages. The use of deep
learning for both supervised and unsupervised applications has increased significantly in recent
years. Convolutional Neural Networks (CNNs) are one of these models that have excelled in object
identification and classification tasks (CNN). The dataset is filtered from MNIST: HAM10000, which
has a sample size of 10015 and includes seven different types of skin lesions. Data preprocessing
methods include sampling, segmentation using an autoencoder and decoder, and dull razor. The model
was trained using transfer learning methods like DenseNet169 and Resnet 50. Different ratios were
used for the training and assessment, including 80:20, 70:30, and 40:60. When undersampling and
oversampling were compared, DenseNet169’s undersampling technique produced accuracy of 91.2% with a
f1-measure of 91.7% and Resnet50’s oversampling technique produced accuracy of 83% with a f1-
measure of 84%. The future extension of this study includes increasing forecast accuracy through
parameter tuning.

DeepMetaForge: A Deep Vision-Transformer Metadata-Fusion

Network for Automatic Skin Lesion Classification
Abstract:
Skin cancer is a dangerous form of cancer that develops slowly in skin cells. Delays in diagnosing
and treating these malignant skin conditions may have serious repercussions. Likewise, early skin
cancer detection has been shown to improve treatment outcomes. This paper proposes DeepMetaForge, a
deep-learning framework for skin cancer detection from metadata-accompanied images. The proposed
framework utilizes BEiT, a vision transformer pre-trained as a masked image modeling task, as the
image-encoding backbone. We further propose merging the encoded metadata with the derived visual
characteristics while compressing the aggregate information simultaneously, simulating how photos
with metadata are interpreted. The experiment results on four public datasets of dermoscopic and
smartphone skin lesion images reveal that the best configuration of our proposed framework yields
87.1% macro-average F1 on average. The empirical scalability analysis further shows that the
proposed framework can be implemented in a variety of machine-learning paradigms, including
applications on low-resource devices and as services. The findings shed light on not only the
possibility of implementing telemedicine solutions for skin cancer on a nationwide scale that could
benefit those in need of quality healthcare but also open doors to many intelligent applications in
medicine where images and metadata are collected together, such as disease detection from CT-scan
images and patients’ expression recognition from facial images.

Conclusion and Future Work

In this experiment, we proposed a novel network architecture, DeepMetaForge, for skin lesion
classification, incorporating image and metadata information to improve classification accuracy.
The proposed architecture features BEiT image-encoding backbone and the novel Deep Metadata Fusion
Module (DMFM) that integrates visual and metadata features while blending them together
simultaneously. We evaluated the performance of the DeepMetaForge network, along with other state-
of-the-art approaches, on four datasets comprising skin lesion images taken from both dermoscopy
and smartphone cameras. The results demonstrated that the proposed network with the BEiT image
encoder backbone not only generalized well to different image sources and metadata compositions but
also outperformed other networks in terms of F1, accuracy, and MCC, making it a suitable model for
skin lesion classification when images and their metadata are available. A scalability analysis was
conducted to investigate how the required computation resources would impact the classification
performance of the proposed approach. This work can be extended by framing the problem as
multiclass classification or object detection tasks, which can have significant implications for
pre-screening skin lesion problems in remote areas. Future work can also focus on adapting the
network to meet the specific needs of remote communities, ultimately improving public healthcare in
underdeveloped and developing countries.