Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion

• Author(s): Shima Ali Sadia; Eusha Mohtasim
• Paper ID: 1719647
• Page: 582-596
• Published Date: 08-07-2026
• Published In: Iconic Research And Engineering Journals
• Publisher: IRE Journals
• e-ISSN: 2456-8880
• Volume/Issue: Volume 10 Issue 1 July-2026
• DOI: https://doi.org/10.64388/IREV10I1-1719647

Abstract

Background: Although skin cancers are the most prevalent cancer diagnosed in the United States, racial and ethnic minority individuals and people residing in rural and other underserved communities are overrepresented with skin cancers diagnosed late, lack timely access to referral to experts, and are more likely to die from skin cancer. While effective in controlled settings machine learning (ML) based diagnostic tools have limited impact on a particular population that would benefit from increased capacity to be triaged due to the datasets clinical biases, geographic barriers, and gaps in referral initiation and completion. Goals: Review and critique the evidence of the use of advanced machine learning models to detect early skin cancer; compile a list of specific barriers that prevent such models from making an equitable impact in underserviced communities in the U.S.; and recommend a holistic triage system for ML-based detection of early skin cancer, teledermatology and Methods: The 20 peer-reviewed literature sources that were reviewed and used for this narrative review included foundational studies on machine learning validation for dermatology, studies examining algorithmic bias and a diversity of skin datasets, dermatology workforce and access epidemiology studies, teledermatology implementation science, and dermatology referral completion outcomes research. Patients were sorted in categories based on the different stages of the lesion detection to confirmed dermatologic evaluation process; focus was maintained on the point where patients were dropped from the process. The results: include new deep learning systems which are promising to match the performance of board-certified dermatologists for diagnosis on curated sets of images and the first primary-care skin cancer system to use AI is now authoritatively approved by regulators in the United States. (Venkatesh et al., 2024) However, dermatology AI models that have been trained to predominantly and almost exclusively classify lighter skin tone images are significantly compromised when was applied to models from darker skin tones and less common presentations. A critical and increasing geographic maldistribution of the dermatology workforce (Feng et al., 2018) coupled with an incomplete triage of patients often fail to complete referral to full definitive in-person evaluations, even after it is successful (Duniphin, 2023; Pagani et al., 2023), especially when crossing geographical boundaries or language/racial barriers. Perspectives: While advanced machine learning models are a potentially promising resource for enhancing the capacity to detect early skin cancer in underserved communities in the U.S. by no means can they be assumed to be similarly distributed. To unlock their potential, however, calls for intentional, concurrent investments in a variety of diverse and representative training data, auditing biases, and structured support to complete referrals, coupled with the need for a coordinated triage pathway, all done and delivered as a whole and not in isolation as technical interventions.

Citations

IRE Journals:
Shima Ali Sadia, Eusha Mohtasim "Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion" Iconic Research And Engineering Journals Volume 10 Issue 1 2026 Page 582-596 https://doi.org/10.64388/IREV10I1-1719647

IEEE:
Shima Ali Sadia, Eusha Mohtasim "Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion" Iconic Research And Engineering Journals, vol. 10, no. 1, Jul. 2026, doi: https://doi.org/10.64388/IREV10I1-1719647

APA:
Shima Ali Sadia, Eusha Mohtasim (2026). Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion. Iconic Research And Engineering Journals, 10(1). doi: https://doi.org/10.64388/IREV10I1-1719647

MLA:
Shima Ali Sadia, Eusha Mohtasim "Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion" Iconic Research And Engineering Journals, vol. 10, no. 1, Jul. 2026. Crossref, https://doi.org/10.64388/IREV10I1-1719647

BibTeX

@article{1719647,
author = {Shima Ali Sadia, Eusha Mohtasim},
title = {Transforming Skin Cancer Triage in Underserved U.S. Communities: An In-Depth Examination of Advanced Machine Learning Models for Early Detection and Dermatology Referral Completion},
journal = {Iconic Research And Engineering Journals},
year = {2026},
volume = {10},
number = {1},
pages = {582-596},
issn = {2456-8880},
url = {https://www.irejournals.com/formatedpaper/1719647.pdf},
abstract = {Background: Although skin cancers are the most prevalent cancer diagnosed in the United States, racial and ethnic minority individuals and people residing in rural and other underserved communities are overrepresented with skin cancers diagnosed late, lack timely access to referral to experts, and are more likely to die from skin cancer. While effective in controlled settings machine learning (ML) based diagnostic tools have limited impact on a particular population that would benefit from increased capacity to be triaged due to the datasets clinical biases, geographic barriers, and gaps in referral initiation and completion. Goals: Review and critique the evidence of the use of advanced machine learning models to detect early skin cancer; compile a list of specific barriers that prevent such models from making an equitable impact in underserviced communities in the U.S.; and recommend a holistic triage system for ML-based detection of early skin cancer, teledermatology and Methods: The 20 peer-reviewed literature sources that were reviewed and used for this narrative review included foundational studies on machine learning validation for dermatology, studies examining algorithmic bias and a diversity of skin datasets, dermatology workforce and access epidemiology studies, teledermatology implementation science, and dermatology referral completion outcomes research. Patients were sorted in categories based on the different stages of the lesion detection to confirmed dermatologic evaluation process; focus was maintained on the point where patients were dropped from the process. The results: include new deep learning systems which are promising to match the performance of board-certified dermatologists for diagnosis on curated sets of images and the first primary-care skin cancer system to use AI is now authoritatively approved by regulators in the United States. (Venkatesh et al., 2024) However, dermatology AI models that have been trained to predominantly and almost exclusively classify lighter skin tone images are significantly compromised when was applied to models from darker skin tones and less common presentations. A critical and increasing geographic maldistribution of the dermatology workforce (Feng et al., 2018) coupled with an incomplete triage of patients often fail to complete referral to full definitive in-person evaluations, even after it is successful (Duniphin, 2023; Pagani et al., 2023), especially when crossing geographical boundaries or language/racial barriers. Perspectives: While advanced machine learning models are a potentially promising resource for enhancing the capacity to detect early skin cancer in underserved communities in the U.S. by no means can they be assumed to be similarly distributed. To unlock their potential, however, calls for intentional, concurrent investments in a variety of diverse and representative training data, auditing biases, and structured support to complete referrals, coupled with the need for a coordinated triage pathway, all done and delivered as a whole and not in isolation as technical interventions.},
month = {July}
}