Artificial Intelligence in Dermoscopy: Enhancing Diagnosis to Distinguish Benign and Malignant Skin Lesions

Abstract

This study presents an innovative application of artificial intelligence (AI) in distinguishing dermoscopy images depicting individuals with benign and malignant skin lesions. Leveraging the collaborative capabilities of Google's platform, the developed model exhibits remarkable efficiency in achieving accurate diagnoses. The model underwent training for a mere one hour and 33 minutes, utilizing Google's servers to render the process both cost-free and carbon-neutral. Utilizing a dataset representative of both benign and malignant cases, the AI model demonstrated commendable performance metrics. Notably, the model achieved an overall accuracy, precision, recall (sensitivity), specificity, and F1 score of 92%. These metrics underscore the model's proficiency in distinguishing between benign and malignant skin lesions. The use of Google's Collaboration platform not only expedited the training process but also exemplified a cost-effective and environmentally sustainable approach. While these findings highlight the potential of AI in dermatopathology, it is crucial to recognize the inherent limitations, including dataset representativity and variations in real-world clinical scenarios. This study contributes to the evolving landscape of AI applications in dermatologic diagnostics, showcasing a promising tool for accurate lesion classification. Further research and validation studies are recommended to enhance the model's robustness and facilitate its integration into clinical practice.

Keywords: artificial intelligence (ai); benign lesions; dermoscopy image analysis; malignant lesions; skin lesions.

Figure 1. CNN model recognizing different images of benign lesions.

(A) Nevus, (B) an ulcer, (C) dermatofibroma, and (D) a mole. The panel of images illustrates distinctive characteristics indicative of a benign skin lesion, such as symmetrical structure, consistent coloration, and scaly appearance. These features serve as crucial cues utilized by the developed AI model for detection and diagnosis. AI, artificial intelligence; CNN, convolutional neural network

Figure 2. CNN model detecting images of different malignant lesions.

(A) Melanoma, (B) squamous cell carcinoma, (C) basal cell carcinoma, and (D) Kaposi sarcoma. The board of images portrays distinctive features, suggesting a malignant skin lesion, which includes an asymmetrical shape, variation in color, and irregular borders. These characteristics serve as vital indicators leveraged by the developed AI model for the detection and diagnosis of malignant skin lesions. AI, artificial intelligence; CNN, convolutional neural network

Figure 3. Precision-recall curve for benign and malignant lesion detection model.

The graphical depiction illustrates the precision and recall of the neural network model across different confidence intervals.

Figure 4. Confusion matrix.

Various metrics, including accuracy, precision, recall (sensitivity), specificity, and F1 score, were computed using the data extracted from the confusion matrix.