A Major Milestone in Diagnosing Skin Cancer with Artificial Intelligence
In this article
Rapid advances in artificial intelligence (AI) are also driving important developments in skin cancer diagnosis. Image recognition is one of AI's most successful fields and plays an important role in early skin cancer diagnosis. AI can identify skin cancer as accurately as dermatologists, as demonstrated using approaches such as deep-learning artificial neural networks.
Skin cancers are among the most common cancers worldwide, and their incidence is rising. About 5% are the aggressive form known as melanoma. Its incidence has also increased rapidly worldwide in recent years, more prominently in women than in men. In particular, among women aged 50 and over, melanoma incidence increased by about 1% annually between 2015 and 2019.
Traditional skin cancer diagnosis is expensive, requires a qualified doctor, and takes time. To help diagnose skin cancer, AI tools and diagnostic support systems are therefore becoming increasingly common.
Many AI applications have been developed for skin cancer diagnosis in recent years. Some initially failed to produce sufficiently accurate results. Over time, however, their accuracy has improved substantially, making them valuable diagnostic tools.
Study Details
AI research presented on 12 October 2023 at the European Academy of Dermatology and Venereology (EADV) Congress 2023 evaluated 22356 patients with suspected skin cancer over a period of 2.5 years.
- Version 3 of the AI application tested in this study achieved 100% sensitivity for melanoma (59/59 cases detected), with reported accuracy rates of 99.5% for all skin cancers (189/190) and 92.5% for precancerous lesions (541/585).
Test sensitivity means a test's ability to detect a particular condition or disease. More specifically, it is the proportion of people who truly have the condition or disease whom the test correctly identifies. A highly sensitive test can correctly detect a large proportion of affected individuals, enabling early diagnosis and treatment. Sensitivity is usually expressed as a percentage and gives healthcare professionals important information about a test's reliability and effectiveness.
- The third version shows a substantial improvement over the first model tested in 2021. That model detected melanoma with 85.9% accuracy (195/227), all skin cancers with 83.8% (903/1078), and precancerous lesions with 54.1% (496/917).
Table 1: Performance of the AI Application by Version
|
AI Version 1 (20 April 2020 - 14 July 2021) |
AI Version 2 (15 July 2021 - 21 April 2022) |
AI Version 3 (22 April 2022 - 5 January 2023) |
|
|
Melanoma sensitivity |
85.9% |
97.1% |
100% |
|
Sensitivity for all skin cancers |
83.8% |
98.5% |
99.5% |
|
Precancerous lesion sensitivity |
54.1% |
94.0% |
92.5% |
|
Benign lesion specificity |
79.7% |
49.4% |
75.3% |
The study's lead author, Dr. Kashini Andrew from the United Kingdom, states: “This study demonstrates how rapidly AI develops and learns, and how high accuracy can be directly attributed to improvements in AI training techniques and the quality of the training data. Between April 2022 and January 2023, the latest software version saved more than 1000 face-to-face doctor appointments in a secondary-care setting, providing more time for patients needing urgent attention.”
Specialist Supervision Is Still Needed
Although the data are very encouraging, the researchers state that AI should not be used as an independent detection tool without the support of a consultant dermatologist. One basal cell carcinoma among the 190 cases was missed and subsequently identified on a second reading by a dermatologist “safety net.” This further illustrates the need for appropriate clinical oversight.
AI's role in dermatology and how best to use it are widely debated. Further research with appropriate clinical oversight may permit its use as a triage tool. Any pathway must, however, demonstrate cost-effectiveness, and AI is currently not a stand-alone tool in dermatology. Studies such as this suggest that it holds considerable promise for future healthcare delivery.
Sources
Was this article helpful?
Your feedback helps us improve our content.
Reader comments
Share your comment
Your name and comment are published only after administrator approval. Your email address is never shown to visitors.
Be the first to comment.