Machine Learning-Based Early Breast Cancer Detection Through Temperature and Color Skin with Non-Invasive Smart Device

Aug 5, 2026

Breast cancer remains a major global health challenge, often resulting in late-stage diagnosis despite the effectiveness of mammograms, ultrasounds, and biopsies. This research explores machine-learning–based early detection using physiological data from a smart, non-invasive wearable. Subtle temperature and skin-color variations are captured in real time and used to classify cancerous conditions. Experiments on data collected from a modified mannequin evaluate Random Forest, SVM, and MLP, with Random Forest achieving the best performance. Future work emphasizes broader datasets, augmentation, transfer learning, and clinical validation.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Introduction

Breast cancer burden and screening disparities

Limitations of common diagnostic techniques

Breast self-examination and early physical indicators

Method Overview

Smart non-invasive device and physiological data capture

Machine Learning Models and Evaluation

Random Forest performance

SVM and MLP results

Future Directions

Generalizability, augmentation, transfer learning, clinical validation

Machine Learning-Based Early Breast Cancer Detection Through Temperature and Color Skin with Non-Invasive Smart Device Sona Regina Salsabila 1, Sugiyarto Surono 1, Irsyadul Ibad2, Eko Prasetyo2, Arsyad Cahya Subrata2,3,4, Aris Thobirin 1 1Department of Mathematic, Universitas Ahmad Dahlan, Yogyakarta 55191, Indonesia 2Department of Electrical Engineering, Universitas Ahmad Dahlan, Yogyakarta 55191, Indonesia 3Center of Electrical Electronic Research & Development, Universitas Ahmad Dahlan, Yogyakarta 55191, Indonesia 4Embedded Systems and Power Electronics Research Group, Universitas Ahmad Dahlan, Yogyakarta 55191, Indonesia | ARTICLE INFO
Article history:
Received November 21, 2024 Revised December 08, 2024 Published December 20, 2024
Keywords:
Breast cancer detection; Physiological data;
Non-invasive diagnostics; Machine Learning; Early screening | ABSTRACT
Breast cancer remains a significant global health issue, affecting millions of women and often leading to late-stage diagnoses. Traditional diagnostic methods, such as mammograms, ultrasounds, and biopsies, are effective but can be costly, invasive, and not widely accessible, causing delays in detection and treatment. This research highlights the potential of using machine learning models with physiological data for early breast cancer detection. By capturing subtle physiological variations from a smart bra, the device allows real-time, non-invasive monitoring, offering a preventive solution that reduces the need for frequent clinical visits. The data were collected from a modified mannequin designed to simulate conditions related to breast cancer. To classify cancerous conditions based on temperature and color data, three machine learning models were evaluated. The Random Forest (RF) model proved to be the most effective, achieving 89% accuracy, 86.11% precision, 88.57% recall, and an F1-score of 87.33%, demonstrating strong performance in identifying complex patterns. The Support Vector Machine (SVM) achieved an accuracy of 81.25%, precision of 85.7%, recall of 80%, and an F1-score of 82.64%. The Multilayer Perceptron (MLP) exhibited an accuracy of 72%, precision of 69.69%, recall of 65.71%, and an F1-score of 67.52%, suggesting potential but requiring further optimization. These models serve as valuable tools to assist medical professionals in early screening efforts. Future research should aim to improve the models’ generalizability by expanding the dataset, utilizing data augmentation, applying transfer learning, and incorporating additional variables. Clinical validation and human trials are essential next steps to evaluate the system's effectiveness. | | --- | --- | | This work is licensed under a Creative Commons Attribution-Share Alike 4.0
| | | Corresponding Author:
Sugiyarto Surono, Department of Mathematic, Universitas Ahmad Dahlan, Yogyakarta 55191, Indonesia Email: [sugiyarto@math.uad.ac.id](sugiyarto@math.uad.ac.id) | | 1. INTRODUCTION Breast cancer remains a global health issue. The World Health Organization (WHO) predicting this trend will continue [1], [2], [3] . Breast cancer is the most common cancer in women, with 2.3 million new cases and 685.000 deaths reported in 2020 [4], [5] . This burden is not distributed equally, as the incidence and mortality rates vary significantly across regions. For instance, developed countries typically report higher incidence rates due to more effective screening programs, while developing countries face higher mortality rates because of late detection and limited access to treatment [6] . Prediction indicates that by 2040, the annual incidence of new cases is expected to rise by over 40% from 2018, surpassing 3 million cases each year [7] . Diagnosing breast cancer is a critical step in the treatment process. Common diagnostic techniques include physical exams, mammograms, ultrasounds, and biopsies [8], [9],[10] . This method has high accuracy to detecting breast cancer b

FAQs

Why are non-invasive early breast cancer detection methods important?

Traditional diagnostics can be costly, invasive, and less accessible, which delays detection. Non-invasive monitoring aims to support earlier screening and reduce dependence on frequent clinical visits.

What physiological signals does the proposed smart device use?

The system captures subtle temperature and skin-color variations from a smart bra to reflect early indicators associated with breast cancer conditions.

Which machine learning model performed best in the study?

Random Forest delivered the strongest results, reaching 89% accuracy with an F1-score of 87.33%, outperforming SVM and MLP on the reported metrics.

Aug 5, 2026

Upload by Theodore