Page 1 of 8
British Journal of Healthcare and Medical Research - Vol. 10, No. 6
Publication Date: December 25, 2023
DOI:10.14738/bjhmr.106.16157.
Secara, I. A. & Hordiiuk, D. (2023). Revolutionizing Dermatology: The Use of Computer Vision for Automated Skin Lesion Analysis.
British Journal of Healthcare and Medical Research, Vol - 10(6). 340-347.
Services for Science and Education – United Kingdom
Revolutionizing Dermatology: The Use of Computer Vision for
Automated Skin Lesion Analysis
Ion-Alexandru Secara
Zen
Dariia Hordiiuk
AiSport
ABSTRACT
The integration of computer vision in dermatology holds promise for
revolutionizing skin lesion analysis. This article offers a comprehensive
exploration of the current challenges in skin lesion analysis, the potential of
computer vision in dermatology, methods and algorithms in skin lesion analysis,
the impact of computer vision on dermatology, its respective limitations and future
directions. Additionally, the article includes a review of existing literature and
original analysis. The potential of computer vision in dermatology is discussed,
emphasizing its capacity to enhance diagnostic accuracy, reduce human error in
dermatological assessments, and provide faster and more accessible analysis.
Furthermore, the potential of computer vision in dermatology extends beyond
diagnostic applications, with the ability to comply with stricter data privacy
protection laws and facilitate the development of visual object detection platforms
powered by federated learning, ensuring data privacy and security in
dermatological applications. The article concludes with a summary of key findings,
implications for the field of dermatology, and recommendations for future
research.
Keywords: computer vision in dermatology, challenges of computer vision in
dermatology, automated skin lesion analysis, revolutionizing dermatology, applications of
computer vision in dermatology
INTRODUCTION
The field of dermatology is undergoing a significant transformation with the integration of
computer vision and machine learning techniques. Recent advancements in access to large
datasets, faster computing, and cheaper data storage have encouraged the development of
machine learning algorithms with human-like intelligence in dermatology [1]. These
technologies have the potential to address the current challenges in skin lesion analysis,
including limitations of traditional methods, issues with accuracy and efficiency in current
practices, and the need for improved diagnostic accuracy [2]. The potential of computer vision
in dermatology is evident in its ability to improve diagnostic accuracy, reduce human error in
dermatological assessments, and provide faster and more accessible analysis [3]. Furthermore,
the use of deep learning models and convolutional neural networks has shown promising
results in the automated detection and classification of skin lesions [4]. The integration of these
technologies has the potential to revolutionize the field of dermatology by enabling remote
Page 2 of 8
341
Secara, I. A. & Hordiiuk, D. (2023). Revolutionizing Dermatology: The Use of Computer Vision for Automated Skin Lesion Analysis. British Journal of
Healthcare and Medical Research, Vol - 10(6). 340-347.
URL: http://dx.doi.org/10.14738/bjhmr.106.16157.
diagnosis and triaging, improving patient outcomes, and enhancing the efficiency of
dermatological assessments [5]. However, the implementation of these technologies also raises
ethical considerations, challenges, and limitations that need to be addressed [6]. This paper
aims to provide a comprehensive overview of the potential of computer vision in dermatology,
the methods and algorithms used in automated skin lesion analysis, and the impact of computer
vision on the field of dermatology. Additionally, it will review existing literature and provide
original insights and contributions to the field, with the goal of guiding future research and
advancements in this rapidly evolving field.
CURRENT CHALLENGES IN SKIN LESION ANALYSIS
Skin lesions present a diverse range of visual characteristics, making their analysis and
classification a complex task. Dermoscopic images from the International Skin Imaging
Collaboration (ISIC) skin lesion classification challenge datasets are available in various sizes,
requiring comprehensive analysis and classification techniques. Additionally, the automatic
segmentation of skin lesions in dermoscopic images is challenging due to artifacts, indistinct
boundaries, low contrast, and varying sizes and shapes of the lesion images. The analysis of skin
lesion images is further complicated by the high intraclass similarity and variance, necessitating
advanced techniques for accurate classification and analysis [7].
Traditional methods for skin lesion analysis face significant limitations, including the difficulty
in differentiating oral melanomas from nevi and melanotic macules, posing a diagnostic
challenge despite morphologic similarities with cutaneous melanomas. Furthermore, the
association between serum periostin levels and the severity of arsenic-induced skin lesions
highlights the complexity of understanding the underlying biological mechanisms of skin
lesions. The imbalanced datasets, low contrast lesions, and the extraction of irrelevant or
redundant features also present challenges in traditional methods for skin lesion analysis [8].
The accuracy and efficiency of current practices in skin lesion analysis are hindered by the
challenges of early detection and accurate identification of skin lesions, which remain
significant obstacles. Furthermore, the assessment of Raman spectroscopy for reducing
unnecessary biopsies for melanoma screening aims to improve accuracy while reducing
invasive procedures, highlighting the need for more efficient diagnostic methods. The
challenges in the development of artificial intelligence (AI) in dermatology also impact the
accuracy and efficiency of current practices, emphasizing the need for advancements in AI tools
for skin disease diagnosis [9].
The field of computer vision has had multiple contributions to the field of dermatology, through
the development of novel algorithms for skin lesion classification, segmentation, and diagnosis.
Recent studies have proposed innovative approaches, such as the use of multi-modal medical
features for skin lesion classification, and the development of lightweight models for skin lesion
detection, highlighting the potential of computer vision in advancing dermatological imaging
and analysis [10]. Additionally, the development of deep learning-based diagnostic systems
with dual-channel image and extracted text data has provided valuable insights into the
integration of artificial intelligence in dermatology, emphasizing the potential of computer
vision in improving diagnostic accuracy and clinical decision-making [11]. The original insights
and contributions to the field have collectively advanced the understanding and application of
Page 3 of 8
342
British Journal of Healthcare and Medical Research (BJHMR) Vol 10, Issue 6, December- 2023
Services for Science and Education – United Kingdom
computer vision in dermatology, paving the way for the development of innovative diagnostic
tools and systems for skin lesion analysis.
THE POTENTIAL OF COMPUTER VISION IN DERMATOLOGY
Computer vision has emerged as a transformative technology in dermatology, offering
promising opportunities to revolutionize the analysis and diagnosis of skin lesions. The
application of computer vision techniques in dermatology has garnered significant attention
due to its potential to enhance diagnostic accuracy, improve patient outcomes, and streamline
dermatological assessments. Recent studies have highlighted the potential of deep learning- enabled medical computer vision in various medical fields, including radiology, with a focus on
X-rays, CT, and MRI [12]. Furthermore, the interpretation of dermatological images has been at
the forefront of computer vision applications in medicine, with recent data suggesting that
artificial neural networks (ANN) may outperform human dermatologists in the diagnosis of
certain dermatological lesions [13].
In the context of surgery, computer vision has demonstrated its potential in enhancing surgical
procedures, and its application in dermatological surgery holds promise for improving
procedural accuracy and patient outcomes [14]. Additionally, the development of checklists for
the evaluation of image-based artificial intelligence reports in dermatology emphasizes the
growing importance of ensuring the reliability and safety of computer vision applications in
dermatological diagnosis, triage, monitoring, segmentation, and decision support [15].
The potential of computer vision in dermatology is further underscored by the significant
advancements in deep learning methods, which have shown great promise in applications such
as image classification and natural language processing [16]. Moreover, the integration of
artificial intelligence in dermatology has paved the way for the development of advanced
diagnostic systems for skin diseases, leveraging dual-channel image and extracted text data for
accurate diagnosis and classification [17].
In the context of computer vision, the application of binocular stereo vision has demonstrated
practical value in various domains, including intelligent monitoring, workpiece measurement,
and three-dimensional reconstruction [18]. Furthermore, the promising results of computer
vision in complex diagnostics, including dermatology, radiology, and pathology, highlight the
potential of this technology in addressing the challenges of skin lesion analysis and diagnosis
[19]. The potential of computer vision in dermatology extends beyond diagnostic applications,
with the capacity to comply with stricter data privacy protection laws and facilitate the
development of visual object detection platforms powered by federated learning, ensuring data
privacy and security in dermatological applications [20]. Moreover, the integration of computer
vision in endoscopic video analysis holds the potential to scale applications for the benefit of a
wider group of dermatological surgeons and patients, emphasizing its clinical value in
dermatological procedures [21].
METHODS AND ALGORITHMS IN AUTOMATED SKIN LESION ANALYSIS
Preprocessing Techniques in Image Analysis
During image acquisition, dermoscopic images may have certain artifacts such as thin/thick
hair, low contrast image resolution, dark spots/bubbles around the infected skin region, and
irregular lesion boundary that ultimately minimize the accuracy of skin lesion detection. To
Page 4 of 8
343
Secara, I. A. & Hordiiuk, D. (2023). Revolutionizing Dermatology: The Use of Computer Vision for Automated Skin Lesion Analysis. British Journal of
Healthcare and Medical Research, Vol - 10(6). 340-347.
URL: http://dx.doi.org/10.14738/bjhmr.106.16157.
handle these challenging tasks, preprocessing helps in accurate detection of the skin lesion. For
instance, research about deep semantic segmentation and multi-class skin lesion classification
[22] utilized the high pass filter for highlighting the edges and subsequently eliminated
additional illumination using a homomorphic filter.
To enhance the precision of skin lesion segmentation and classification, the research mentioned
above employed a Contrast Enhancement technique that combines the haze reduction
approach with the fast local Laplacian filters method [23]. This fusion process is executed
through the HSV color transformation. This approach was utilized in a manner analogous to the
diagnosis of skin cancer.
In order to discriminate between benign and malignant skin lesions three standard pre- processing steps for transfer learning used [24]:
• Normalization of the images by subtracting the mean RGB value of the ImageNet dataset.
• Resizing the images using bicubic interpolation to be fed to the networks (227x227 and
224x224).
• Augmentation of the training set by rotating the images by 0, 90, 180 and 270 degrees
and then further applying horizontal flipping.
These methods play a vital role in achieving optimal accuracy when it comes to computer- vision-based skin lesion detection and classification.
Feature Extraction Methods for Skin Lesion Classification
Feature extraction is a critical step in automated skin lesion analysis. For instance, [24] used
AlexNet, VGG16 and ResNet-18 as optimized feature extractors. The study [25] utilized NasNet
as a deep feature extractor followed by feature selection techniques as HWOA (Harmony Whale
Optimization Algorithm) and EMI (Electromagnetic Induction). The survey [26] mentions
several feature extraction techniques used in the analysis of skin lesions. These include:
1. Pattern Analysis: This involves examining the size and distribution of patterns visible in
dermoscopy images. The patterns are divided into local and global patterns. Global
patterns are textural structures and local patterns are the irregular or regular structure
to define whether a skin lesion is malignant or benign.
2. Shape Features: The shape features of the lesions are examined by dividing the region
into two sub-regions. Geometrical measures from segmented lesion regions are
computed to assess the border irregularity.
3. Color Variation: RGB color space is used for skin lesion representation. The standard
deviation, skewness, and variance are computed for each color channel for the lesion
regions.
4. Textural Analysis: This is used to assist in discriminating between malignant and benign
skin lesions. The features are also extracted based on histograms based on the intensity
and gradient histogram to represent the texture features.
5. Deep Residual Network: This is used to overcome issues based on lower exactness by
making a plain neural network with a more profound layer.
6. Convolutional and Max Pooling Layers: These are used for down-sampling and feature
extraction.