ARTIFICIAL INTELLIGENCE–ASSISTED DIAGNOSIS FOR EARLY MYCOSIS FUNGOIDES DETECTION IN SKIN BIOPSIES
- 1Universidade Federal Fluminense, Rio de Janeiro, RJ, Brazil
Abstract
Background: Mycosis fungoides (MF) is the most common cutaneous T-cell lymphoma. Diagnosis in the patch stage is a clinical and histopathological challenge, as changes are subtle and mimic benign inflammatory dermatoses. Aim: To identify the most relevant immunohistochemical, histological, and clinical variables for MF diagnosis, with emphasis on the patch stage, using statistical methods and machine learning algorithms. Material and Methods: Cross-sectional study with 94 skin biopsies, including 40 MF cases (10 in patch stage) and 54 controls with alternative diagnoses. Thirty-five variables were analyzed (CD2, CD3, CD4, CD5, CD7, CD8, and CD20 markers in epidermis and dermis; binary histological variables; clinical data). Feature selection used five methods: Mutual Information, Random Forest (Gini), L1 Logistic Regression (Lasso), Gradient Boosting, and Permutation Importance. Predictive performance was evaluated by stratified 5-fold cross-validation. Results: The most relevant variables by consolidated ranking were: age (mean rank 4.0; p=0.023), loss of CD7 expression in the epidermis (rank 8.8; p<0.001; r=0.62) and dermis (rank 9.4; p<0.001), CD5 in the dermis (p=0.013), and skin color (p=0.036). For MF-patch specifically, female sex (p=0.005), dermal CD5 (p=0.013; r=0.305), and disease duration in months (p=0.044; r=0.315) were highlighted. Gradient Boosting achieved the best predictive performance (AUC-ROC=0.861; Balanced Accuracy=0.788). Conclusion: Loss of CD7 and CD5 expression constitutes the most discriminative immunohistochemical profile for MF. Machine learning algorithms demonstrated high discriminatory capacity, representing a potential auxiliary tool for dermatopathological diagnosis. All the authors declare no potential conflict of interest. Ethical approval and informed consent are registered under the CAAE number 78754624.8.0000.5243. The authors provided the research funding
Related
- Artificial intelligence-assisted bronchoscopy for the diagnosis and treatment of pulmonary diseases: Advances and clinical practice
- Diagnostic accuracy of tongue diagnosis for stroke syndrome differentiation in traditional Chinese medicine: a systematic review, meta-analysis, and narrative synthesis of artificial intelligence-assisted models.
- Artificial intelligence-assisted training for proximal caries diagnosis by undergraduate dental students: a randomized experimental study.
- Effectiveness of artificial intelligence-assisted colonoscopy for colorectal lesion detection: a systematic review and meta-analysis of randomized controlled trials
- Generative Artificial Intelligence Models in Breast Cancer Diagnosis: A Systematic Review
BibTeX
@article{pereira2026artificial,
title = {{ARTIFICIAL INTELLIGENCE–ASSISTED DIAGNOSIS FOR EARLY MYCOSIS FUNGOIDES DETECTION IN SKIN BIOPSIES}},
author = {Miguel Augusto Martins PEREIRA and Luciana PANTALEÃO},
journal = {Hematology, Transfusion and Cell Therapy},
year = {2026},
month = oct,
doi = {10.1016/j.htct.2026.107735},
url = {https://doi.org/10.1016/j.htct.2026.107735},
note = {CC BY},
}
RIS
TY - JOUR
TI - ARTIFICIAL INTELLIGENCE–ASSISTED DIAGNOSIS FOR EARLY MYCOSIS FUNGOIDES DETECTION IN SKIN BIOPSIES
AU - Miguel Augusto Martins PEREIRA
AU - Luciana PANTALEÃO
PY - 2026
DA - 2026/10/23
JO - Hematology, Transfusion and Cell Therapy
DO - 10.1016/j.htct.2026.107735
UR - https://doi.org/10.1016/j.htct.2026.107735
AB - Background: Mycosis fungoides (MF) is the most common cutaneous T-cell lymphoma. Diagnosis in the patch stage is a clinical and histopathological challenge, as changes are subtle and mimic benign inflammatory dermatoses. Aim: To identify the most relevant immunohistochemical, histological, and clinical variables for MF diagnosis, with emphasis on the patch stage, using statistical methods and machine learning algorithms. Material and Methods: Cross-sectional study with 94 skin biopsies, including 40 MF cases (10 in patch stage) and 54 controls with alternative diagnoses. Thirty-five variables were analyzed (CD2, CD3, CD4, CD5, CD7, CD8, and CD20 markers in epidermis and dermis; binary histological variables; clinical data). Feature selection used five methods: Mutual Information, Random Forest (Gini), L1 Logistic Regression (Lasso), Gradient Boosting, and Permutation Importance. Predictive performance was evaluated by stratified 5-fold cross-validation. Results: The most relevant variables by consolidated ranking were: age (mean rank 4.0; p=0.023), loss of CD7 expression in the epidermis (rank 8.8; p<0.001; r=0.62) and dermis (rank 9.4; p<0.001), CD5 in the dermis (p=0.013), and skin color (p=0.036). For MF-patch specifically, female sex (p=0.005), dermal CD5 (p=0.013; r=0.305), and disease duration in months (p=0.044; r=0.315) were highlighted. Gradient Boosting achieved the best predictive performance (AUC-ROC=0.861; Balanced Accuracy=0.788). Conclusion: Loss of CD7 and CD5 expression constitutes the most discriminative immunohistochemical profile for MF. Machine learning algorithms demonstrated high discriminatory capacity, representing a potential auxiliary tool for dermatopathological diagnosis. All the authors declare no potential conflict of interest. Ethical approval and informed consent are registered under the CAAE number 78754624.8.0000.5243. The authors provided the research funding
SN - 2531-1379
SN - 2531-1387
ER -