Please login or create an account. If you do not have access to this content, you will be shown a 30 second preview and licensing options.
- Presentation
Augmented Intelligence in Dermatopathology: Opportunities, Applications, and Challenges
Description
The speaker, a dermatologist and dermatopathologist at Mayo Clinic, describes how her fully digital workflow reflects the broader shift from light microscopy to digital pathology and AI. She outlines the historical roots of computer-assisted dermatopathology and summarizes current attitudes from a 2020 survey showing both enthusiasm and fear about AI replacing physicians. The talk organizes AI use into four main goals: detection, characterization, quantification, and prediction. Examples include detecting melanoma metastases in sentinel lymph nodes, identifying nevoid melanoma, classifying common skin lesions, triaging cases, analyzing inflammatory and immunofluorescence studies, counting mitoses and PRAME staining, assessing tumor-infiltrating lymphocytes, and predicting prognosis, BRAF mutation status, immunotherapy response, and sentinel node positivity from H&E slides. She also discusses generative AI for report drafting, image synthesis, and multimodal learning that combines pathology, EMR, radiology, and genomic data. A key message is that dermatopathologists must remain involved in algorithm development and validation. The speaker emphasizes major challenges such as low-power versus pixel-level interpretation differences, lack of ground truth, outdated training labels, automation bias, hallucinations, cost, scanning artifacts, and workflow adoption, concluding that AI can augment but not replace expert clinical judgment.
View moreConclusions
- Augmented intelligence has strong potential to improve dermatopathology by assisting with detection, characterization, quantification, and prognosis prediction.
- The most successful current applications appear to be in narrow, well-defined tasks such as finding metastases, triaging lesions, and counting or quantifying histologic features.
- AI can increase efficiency and may help identify subtle lesions or mimics that could otherwise be missed, but its performance is still imperfect in more complex or rare diagnoses.
- For inflammatory and immunodermatologic disease, AI shows promise, but the evidence base is still comparatively limited.
- AI-derived quantification may make subjective measures like mitotic count, PRAME scoring, and tumor-infiltrating lymphocytes more reproducible.
- Models may also help predict clinically important outcomes such as mutation status, treatment response, and sentinel lymph node positivity from routine histology.
- Generative AI and large language models could eventually support report drafting, image synthesis, and multimodal clinical decision-making.
- Dermatopathologists should remain actively involved in building, curating, and supervising AI systems because expert input improves training quality and performance.
- A major risk is automation bias, so AI should augment rather than replace human judgment.
- Important barriers to adoption include lack of ground truth, changing disease definitions over time, cost, scanner and image-quality limitations, and the need for workflow adaptation.
- A pathology foundation model for cancer diagnosis and prognosis prediction
- Computerized dermatopathologic diagnosis / Brian Potter and Salve G. Ronan#10.1016/s0190-9622(87)70183-2
- Attitudes Toward Artificial Intelligence Within Dermatopathology: An International Online Survey#10.3389/fmed.2020.591952
- Making Pathologists Ready for the New Artificial Intelligence Era: Changes in Required Competencies#10.1016/j.modpat.2024.100657
- Deep learning detection of melanoma metastases in lymph nodes#10.1016/j.ejca.2023.04.023
- Artificial Intelligence Applied to a First Screening of Naevoid Melanoma: A New Use of Fast Random Forest Algorithm in Dermatopathology#10.3390/curroncol30070452
- Histologic Screening of Malignant Melanoma, Spitz, Dermal and Junctional Melanocytic Nevi Using a Deep Learning Model#10.1097/dad.0000000000002232
- Machine learning techniques for mitoses classification
- Digital quantification of PRAME for distinguishing melanoma from nevi compared to manual assessment#10.1016/j.prp.2024.155543
- Artificial Intelligence-Detected Tumor-Infiltrating Lymphocytes and Outcomes in Anti-PD-1-Based Treated Melanoma#10.1001/jamaoncol.2025.4072
- Machine Learning Techniques in Predicting BRAF Mutation Status in Cutaneous Melanoma From Clinical and Histopathologic Features#10.1097/pai.0000000000001075
- Using deep learning to predict anti-PD-1 response in melanoma and lung cancer patients from histopathology images#10.1016/j.tranon.2020.100921
- Deep learning approach to predict sentinel lymph node status directly from routine histology of primary melanoma tumours#10.1016/j.ejca.2021.05.026
- A multimodal generative AI copilot for human pathology
- Classification of Melanocytic Lesions in Selected and Whole-Slide Images via Convolutional Neural Networks#10.4103/jpi.jpi_32_18
- Deep Learning Assisted Diagnosis of Onychomycosis on Whole-Slide Images#10.3390/jof8090912
- Vos et al.
- Neimy et al.
- Stiff et al.
- Rashidi et al.