In a groundbreaking doctoral study, artificial intelligence (AI) has demonstrated its exceptional ability to accurately distinguish malignant skin lesions from benign ones using hyperspectral imaging (HSI). This innovative approach, led by medical licentiate Vivian Lindholm, combines cutting-edge technology with AI analysis to potentially revolutionize skin cancer diagnosis and treatment.
The rising challenge of skin cancer
Skin cancers are among the most prevalent and rapidly increasing forms of cancer globally, including in Finland. Traditional methods for detecting skin cancers often require specialized training for image analysis, making early diagnosis and accurate differentiation between benign and malignant lesions challenging.
Lindholm’s research harnessed the power of hyperspectral imaging, a non-invasive technique that captures a broad range of information about different light wavelengths compared to conventional cameras. This hyperspectral camera provided unique spectral data, akin to a “fingerprint,” for each skin lesion. Complementing this data was a 3D model depicting the skin lesion’s surface structure. AI analysis, employing neural networks, processed this comprehensive data to provide an accurate diagnosis for each skin lesion.
Lindholm’s study assessed the accuracy of this method in differentiating various skin conditions, including melanoma, keratinocyte skin cancer, and precancerous skin changes, from benign lesions. The study included 172 skin lesions, and the results were remarkable. The AI analysis following hyperspectral imaging achieved a 95% accuracy and sensitivity for melanoma, 85-100% sensitivity, 92-100% accuracy for keratinocyte skin cancer, and 84% sensitivity and 94% accuracy for Bowen’s disease.
Mapping skin lesions for precise treatment
Beyond accurate diagnosis, this innovative approach provides mapped images of the skin lesions. This mapping allows for a more precise delineation of lesion borders, particularly beneficial before surgical treatment. By enhancing the accuracy of lesion delineation, this technology has the potential to assist doctors in pinpointing skin lesions that require treatment more accurately, thus reducing the need for unnecessary biopsies and enabling the early detection of skin cancers.
While these findings are promising, Lindholm emphasizes the need for further research and larger datasets encompassing all types of skin lesions to validate and refine the method. This cautious approach ensures that the potential of this technology is fully realized and that it becomes a valuable tool in the fight against skin cancer.
A new dawn for precancerous skin changes
Skin cancer diagnosis, Lindholm’s research also explored treatments for precancerous skin changes, known as actinic keratoses. Traditional treatments, such as photodynamic therapy (PDT), can be effective but often painful. Lindholm’s study introduced two new laser-assisted PDT treatments for actinic keratoses on the head, offering hope for improved patient experiences.
The most significant benefit was observed when using fractional laser before simulated or natural daylight PDT. Fractional laser creates tiny holes in the skin’s surface, significantly enhancing the efficacy of daylight PDT. With this combined treatment, 86% of the lesions improved, while using daylight PDT alone resulted in a 70% improvement. This approach also demonstrated outstanding results for thick lesions, surpassing the effectiveness of standard PDT.
Future optimization and advancements
However, Lindholm underscores the need for additional research to optimize laser settings, enhancing both treatment efficacy and tolerance. These findings suggest that the combination of fractional laser and PDT could become the primary choice for patients with extensive or thick actinic keratoses, offering a less painful yet highly effective treatment option.
Hyperspectral imaging’s superiority lies in its ability to capture a wealth of data about different light wavelengths, far beyond the capabilities of conventional photography. This data provides comprehensive spectral information, allowing for a detailed analysis of the composition of the imaged skin area.
The unique spectrum of each skin lesion primarily depends on chromophores like skin melanin, hemoglobin, bilirubin, water, and fat content. The vast amount of data generated by hyperspectral imaging has prompted recent studies to leverage AI methods for analysis.