# Predicting sentinel lymph node metastasis in melanoma patients: A machine learning-based predictive model

## TL;DR

This paper develops a machine learning model to predict sentinel lymph node metastasis in melanoma patients, aiming to improve clinical diagnosis and reduce patient suffering.

## Contribution

A robust and interpretable machine learning model with a web-based tool for predicting sentinel lymph node metastasis in melanoma.

## Key findings

- A neural network model achieved an F1-score of 0.73 for predicting sentinel lymph node metastasis.
- Key predictive factors include Breslow thickness, microsatellites, Ki67 index, and subtype.

## Abstract

Risk factors for sentinel lymph node (SLN) metastasis in melanoma have been studied. However, there remains a lack of widely applicable models with considerable predictive potential for clinical use.

To developed a well-performing machine learning-based model for predicting SLN metastasis in melanoma patients.

This study collected data on 351 melanoma patients with sentinel lymph node biopsy from our center. Univariate and multivariate logistic regression was used for recognizing key features. The optimal model was selected from 10 machine learning algorithms based on the F1 score. SHapley Additive exPlanations was employed to interpret the outcome of the predictive model. R package Shiny was used to develop a web tool.

The neural network model was chosen with the highest F1-score (0.73), indicating considerable predictive accuracy and calibration. SHapley Additive exPlanations results indicate the related factors for SLN metastasis in melanoma patients were Breslow thickness, microsatellites, Ki67 index, and subtype. Ultimately, we developed a web-based tool to promote the clinical application of the model.

Retrospective study, single institution.

This study established a robust and interpretable machine learning approach for melanoma SLN metastasis prediction. With high sensitivity and accuracy, this approach could reduce misdiagnosis rates and alleviate patient suffering.