HISTOPATHOLOGICAL FEATURE SELECTION IN MYCOSIS FUNGOIDES: AN APPROACH TOWARDS A DIAGNOSTIC MODEL
- 1Universidade Federal Fluminense, Niterói, RJ, Brazil
Abstract
Background: Mycosis fungoides (MF) is the most common subtype of primary cutaneous lymphomas. Early-stage MF (patch or macular stage) remains diagnostically challenging from a dermatopathological perspective, due to overlap with benign dermatoses and clinicopathological discordance. The diagnosis combines clinical, molecular, and histological characteristics. However, the histopathological component of this workup remains an area where standardization and simplification are still lacking. We hypothesize that a streamlined model based on histopathological variables can distinguish early-stage MF (EMF) from more advanced disease and serve as the foundation for a future algorithm to distinguish MF from benign dermatoses. Aim: Identify and rank the histopathological features most informative for recognizing the EMF stage — the phase most susceptible to diagnostic delay and develop a predictive model for EMF using data-driven methods. Material and methods: This exploratory cross-sectional study is based on the review of 40 skin biopsies of confirmed MF over 5 years at a quaternary university hospital. Feature selection for the algorithm was performed using filter methods, notably SelectKBest and Maximum Relevance — Minimum Redundancy (mRMR), applied to a dataset of 45 histopathological variables. Results: The 40 skin biopsies from 39 patients (46% male, 54% female) across different disease stages (patch/macula, plaque, and tumor) were submitted to two different filter methods that consistently selected the same five histopathological variables as most informative for the EMF stage: parakeratosis, eccrine involvement, eosinophil presence, granulomatous infiltrate, and perivascular involvement. A predictive model incorporating those five most relevant variables was developed to predict the diagnosis of the early (patch) stage, achieving 87.2% accuracy, with both sensitivity and specificity above 87%. Notably, eosinophils were entirely absent in all patch-stage biopsies, and a strong negative correlation was found between eosinophils and patch stage (r = −0.32, p = 0.047), as well as between parakeratosis and patch stage (r = −0.59, p = 0.001). The absence of eosinophils in early lesions could be useful for distinguishing MF from inflammatory dermatoses—emerging here as a relevant feature that warrants revisiting and further systematic exploration. Conclusion: This study successfully identified a minimal, high-performing set of 5 histopathological features characteristic of EMF and created a streamlined model capable of distinguishing EMF from advanced disease. The consistent absence of eosinophils in the patch stage may represent a potentially underutilized feature. These findings provide a structured basis for the future development of a histology-based algorithm designed to differentiate early MF from benign dermatoses — an objective that remains to be addressed in subsequent studies with appropriate comparator groups.
Related
- Mycosis Fungoides Characterized by Palmar-Plantar Keratosis and Exudation.
- Comparative Outcomes of Single- and Split-Fraction Low-Dose Radiation in Mycosis Fungoides
- Dermoscopic clues for differentiating mycosis fungoides palmaris et plantaris from palmoplantar eczema.
- Trichoscopy of Scalp Folliculotropic Mycosis Fungoides: A Review of the Literature.
- ARTIFICIAL INTELLIGENCE–ASSISTED DIAGNOSIS FOR EARLY MYCOSIS FUNGOIDES DETECTION IN SKIN BIOPSIES
BibTeX
@article{pereira2026histopathological,
title = {{HISTOPATHOLOGICAL FEATURE SELECTION IN MYCOSIS FUNGOIDES: AN APPROACH TOWARDS A DIAGNOSTIC MODEL}},
author = {Miguel Augusto Martins PEREIRA and Adriano LIMA E SOUZA and Fernanda RIZZO and Mayra Carrijo ROCHAEL and Luciana PANTALEÃO},
journal = {Hematology, Transfusion and Cell Therapy},
year = {2026},
month = oct,
doi = {10.1016/j.htct.2026.107761},
url = {https://doi.org/10.1016/j.htct.2026.107761},
note = {CC BY},
}
RIS
TY - JOUR
TI - HISTOPATHOLOGICAL FEATURE SELECTION IN MYCOSIS FUNGOIDES: AN APPROACH TOWARDS A DIAGNOSTIC MODEL
AU - Miguel Augusto Martins PEREIRA
AU - Adriano LIMA E SOUZA
AU - Fernanda RIZZO
AU - Mayra Carrijo ROCHAEL
AU - Luciana PANTALEÃO
PY - 2026
DA - 2026/10/23
JO - Hematology, Transfusion and Cell Therapy
DO - 10.1016/j.htct.2026.107761
UR - https://doi.org/10.1016/j.htct.2026.107761
AB - Background: Mycosis fungoides (MF) is the most common subtype of primary cutaneous lymphomas. Early-stage MF (patch or macular stage) remains diagnostically challenging from a dermatopathological perspective, due to overlap with benign dermatoses and clinicopathological discordance. The diagnosis combines clinical, molecular, and histological characteristics. However, the histopathological component of this workup remains an area where standardization and simplification are still lacking. We hypothesize that a streamlined model based on histopathological variables can distinguish early-stage MF (EMF) from more advanced disease and serve as the foundation for a future algorithm to distinguish MF from benign dermatoses. Aim: Identify and rank the histopathological features most informative for recognizing the EMF stage — the phase most susceptible to diagnostic delay and develop a predictive model for EMF using data-driven methods. Material and methods: This exploratory cross-sectional study is based on the review of 40 skin biopsies of confirmed MF over 5 years at a quaternary university hospital. Feature selection for the algorithm was performed using filter methods, notably SelectKBest and Maximum Relevance — Minimum Redundancy (mRMR), applied to a dataset of 45 histopathological variables. Results: The 40 skin biopsies from 39 patients (46% male, 54% female) across different disease stages (patch/macula, plaque, and tumor) were submitted to two different filter methods that consistently selected the same five histopathological variables as most informative for the EMF stage: parakeratosis, eccrine involvement, eosinophil presence, granulomatous infiltrate, and perivascular involvement. A predictive model incorporating those five most relevant variables was developed to predict the diagnosis of the early (patch) stage, achieving 87.2% accuracy, with both sensitivity and specificity above 87%. Notably, eosinophils were entirely absent in all patch-stage biopsies, and a strong negative correlation was found between eosinophils and patch stage (r = −0.32, p = 0.047), as well as between parakeratosis and patch stage (r = −0.59, p = 0.001). The absence of eosinophils in early lesions could be useful for distinguishing MF from inflammatory dermatoses—emerging here as a relevant feature that warrants revisiting and further systematic exploration. Conclusion: This study successfully identified a minimal, high-performing set of 5 histopathological features characteristic of EMF and created a streamlined model capable of distinguishing EMF from advanced disease. The consistent absence of eosinophils in the patch stage may represent a potentially underutilized feature. These findings provide a structured basis for the future development of a histology-based algorithm designed to differentiate early MF from benign dermatoses — an objective that remains to be addressed in subsequent studies with appropriate comparator groups.
SN - 2531-1379
SN - 2531-1387
ER -