top of page

Artificial
Intelligence

Overview

One of the most promising applications of AI is in diagnostic imaging, especially for skin cancer detection. Diagnoses in this field often depend on the subjective visual interpretation of clinical and dermoscopic images, making it an ideal candidate for AI applications.

 

Artificial intelligence (AI) involves creating intelligent machines capable of performing tasks autonomously. Machine learning (ML), a branch of AI, focuses on enabling computers to learn tasks without explicit programming. Modern ML techniques use extensive datasets to identify patterns for classification, showing significant potential in various domains. Among these techniques, deep neural networks, particularly convolutional neural networks (CNNs), have gained prominence due to their superior representation and classification capabilities. CNNs are specialized neural networks designed for image analysis and are typically trained using supervised learning, where images in the training dataset are labeled with diagnoses, serving as the ground truth. These labels are crucial for the system to learn the relationship between input data and diagnoses. (1)

 

In 2017, Esteva et al. demonstrated the effectiveness of CNNs for image-based classification in dermatology (2). Their novel system, using deep CNNs trained on a large dataset of nearly 130,000 clinical images, including 3,000 dermoscopic images, was evaluated for its ability to differentiate between keratinocyte carcinomas and seborrheic keratosis, as well as melanomas and nevi. The CNN's performance was on par with that of human experts. Subsequent studies by Haenssle et al. (2020) and Tschandl et al. (2019b) confirmed these findings across various image sets and skin disease categories. A meta-analysis of 70 studies on automated systems for diagnosing pigmented skin lesions reported an overall sensitivity of 74% (95% CI 66-80%) and specificity of 84% (95% CI 79-88%) for melanoma diagnosis, comparable to dermatologists' performance. (3)

 

In the domain of identifying changing or newly emerging lesions on whole-body photographs, AI has already seen widespread utilization. However, the consensus is that AI will complement rather than replace physicians, serving as a tool to support rather than replace human interaction, especially in sensitive areas like cancer diagnosis. (4)

In areas such as aiding nonexperts in diagnosing and treating skin lesions, AI’s potential is largely untapped but promising.

 

Advantages of AI-Assisted Diagnosis
  • Speed: AI can quickly process and analyze large volumes of images.

  • AI could improve access to specialist-level expertise, especially in regions experiencing a shortage of dermatologists and long waiting times for specialist appointments.

  • AI-based systems are expected to provide higher accuracy compared to human experts.

  • Consistency: AI algorithms provide consistent results regardless of external factors like fatigue or subjective bias, which can affect human dermatologists.

  • AI can act as a supplementary tool for dermatologists, providing a second opinion or flagging suspicious lesions for further review

  • AI systems can also serve as educational resources, helping medical students and practitioners.

  • In the long run, AI can potentially reduce healthcare costs.

 
References:

(1). https://link.springer.com/article/10.1007/s13671-019-00267-0

(2). https://pubmed.ncbi.nlm.nih.gov/28117445/

(3). https://jamanetwork.com/journals/jamadermatology/fullarticle/2736374

(4). https://pubmed.ncbi.nlm.nih.gov/37978982/

(5). https://www.sciencedirect.com/science/article/pii/S0010482520303966

Challenges and Future Prospects

The development and implementation of AI in skin cancer screening face several key challenges. One significant challenge is generalizability. Many image databases concentrate on a limited range of diagnoses and skin types, which can reduce an AI model's effectiveness in real-world clinical settings. These databases often lack comprehensive representation of various lesions, including less common or benign types. Additionally, there may be a bias toward patients who are more tech-savvy or have better access to healthcare facilities equipped with AI technology..

 

Another significant challenge is the scarcity of real-world validation studies to confirm the practical utility of AI and machine learning. Most research has been conducted in controlled experimental environments, underscoring the need for prospective studies in actual clinical settings to ensure that AI models are applicable and effective in practice.

 

Privacy concerns also arise from the handling and storage of sensitive patient data, including images and medical history. Most open-access databases are limited in the number of images they provide due to copyright and privacy issues. These databases often focus primarily on dermoscopic images and may lack clinical metadata and full-body photos, which could enhance diagnostic accuracy by providing additional context. Variability in imaging conditions, such as lighting and camera settings, along with the absence of standardized protocols, adds complexity to the situation. Additionally, the opacity of machine-learning algorithms can make it difficult for clinicians to understand the reasoning behind a diagnosis, potentially impacting trust and accountability.

 

International Skin Imaging Collaboration (ISIC)

Effective AI development requires extensive datasets. The ISIC Archive was created to address this. Sponsored by the International Society for Digital Imaging of the Skin, the International Skin Imaging Collaboration (ISIC) is a joint effort between academia and industry aimed at enhancing melanoma diagnoses and reducing mortality through digital imaging technologies. ISIC provides a public database for benchmarking machine learning algorithms and hosts public challenges.

 

ISIC activities are divided into two main areas:

  1. ISIC Working Groups: These expert groups develop standards for skin imaging, including imaging technologies, techniques, terminology, and metadata.

  2. ISIC Archive: ISIC maintains the largest publicly available database of skin lesion images, with over 40,000 images (primarily dermoscopic, but also clinical) of around 30 different skin lesions, each tagged with a diagnosis.

     

The challenges include three steps: (1) segmentation of the lesion from the background of the images; (2) detection of different dermoscopic features; and (3) classification of the lesion in the image. Each year, the challenges have become more complicated with more images and more diagnoses that are included. ISIC challenges have helped dramatically to increase the amount of research and publications related to AI and skin lesion diagnosis. Dozens of papers describing the different algorithms that were used in the challenges were published following each challenge.

 

The most recent ISIC Challenge, the ISIC 2024 Challenge, was hosted on Kaggle during the summer of 2024. This challenge featured a dataset from thousands of patients across three continents, including 400,000 skin lesion image crops, which resemble cell phone photos and are extracted from Three-Dimensional Total Body Photography (3D TBP). The competition tasked participants with developing AI algorithms to differentiate between histologically confirmed malignant skin lesions and benign ones.

 

References:

(1). https://link.springer.com/article/10.1007/s13671-019-00267-0

(2). https://pubmed.ncbi.nlm.nih.gov/28117445/

(3). https://jamanetwork.com/journals/jamadermatology/fullarticle/2736374

(4). https://pubmed.ncbi.nlm.nih.gov/37978982/

(5). https://www.sciencedirect.com/science/article/pii/S0010482520303966

bottom of page