Feature-based detection of breast cancer using convolutional neural network and feature engineering

Sep 21, 2026

Uploaded by Miles

Breast cancer detection is addressed through a multi-stage machine-learning framework that builds input for a ResNet50 neural network from engineered biomarker-derived features. The approach transforms original feature values into normalized representations using the membership of each feature in a Gaussian distribution estimated from healthy and breast-cancer samples. Evaluation uses the Coimbra breast cancer dataset and the Wisconsin dataset, showing accuracy improvements with 100% and 100% results respectively. Comparisons with prior literature support the method’s reliability, highlighting how normalization reduces misclassifications by improving generality.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Background and motivation

Biomarkers and non-invasive detection

Datasets used for evaluation

Proposed multi-stage methodology

Feature computation and normalization

ResNet50-based classification

Experimental results and validation

Performance on Coimbra dataset

Performance on Wisconsin dataset

Comparison with existing studies

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN Feature-based detection of breast cancer using convolutional neural network and feature engineering Hiba Allah Essa1􀀍, Ebrahim Ismaiel2,3􀀍 & Mhd FirasAl Hinnawi1 Breast cancer (BC) is a prominent cause of female mortality on a global scale. Recently, there has been growing interest in utilizing blood and tissue-based biomarkers to detect and diagnose BC, as this method offers a non-invasive approach. To improve the classification and prediction of BC using large biomarker datasets, several machine-learning techniques have been proposed. In this paper, we present a multi-stage approach that consists of computing new features and then sorting them into an input image for the ResNet50 neural network. The method involves transforming the original values into normalized values based on their membership in the Gaussian distribution of healthy and BC samples of each feature. To test the effectiveness of our proposed approach, we employed the Coimbra and Wisconsin datasets. The results demonstrate efficient performance improvement, with an accuracy of 100% and 100% using the Coimbra and Wisconsin datasets, respectively. Furthermore, the comparison with existing literature validates the reliability and effectiveness of our methodology, where the normalized value can reduce the misclassified samples of ML techniques because of its generality. Keywords Breast cancer, Biomarkers, Convolutional neural networks, Feature engineering, Gaussian distribution Breast cancer (BC) is the leading cause of death among women worldwide after lung cancer, with fatalities of 685,000 in 20201. Currently, mammography is the gold standard and a tool that has the ability to detect breast cancer in its early stages. Despite the efficiency of mammography in detecting and diagnosing cancer, this method for low-or intermediate-risk women under 50 is subject to intense debate. Proponents argue for lower age screening, highlighting potential benefits like increased survival rates, improved workforce participation, and reduced treatment costs2. Conversely, opponents raise concerns about the side effects that may occur in the patient’s body due to frequent exposure to the radiation3. The concept of biomarkers, cancerous markers, bio-compounds, and physical indicators in the human body has prompted researchers and clinicians to focus on identifying all types of cancers and malignant tumors, including breast cancer, as well as how to categorize pathological conditions in general and cancerous conditions in particular4. Several studies have identified some molecular compounds with concentrations that differ between pathological and cancerous cases. These dysregulated concentrations have been categorized as helpful biomarkers for distinguishing between cancerous and healthy (H) cases. Haptoglobin, osteopontin (OPN), carcinoma antigen (CA) 15-3, CA125, and CA19-95 can all be detected using biomarkers like blood-based analysis, which are non-invasive and quantifiable features. Machine learning (ML) tools are therefore urgently needed to reach the necessary level of cancer diagnosis and prediction using available non-invasive biomarkers, as a result of the large data, which is typically close to healthy and cancer levels6,7. The Coimbra breast cancer dataset (CBCD)8 is one of the most important datasets used to investigate the biomarker-based diagnosis and detection of BC using machine learning. The Coimbra dataset contains several biomarkers gathered from routine blood analysis such as insulin, leptin and glucose. Support Vector Machine (SVM) models employing resistin, BMI, glucose, and age exhibit the best performance with an average specificity of 87.1%, sensitivity of 84.85%, and an area under the curve (AUC) of 0.888. Silva et al.9 used a fuzzy neural network method for breast cancer detection with CBCD and reached the best accuracy of 81.04% and a sensitivity of 81.93% using res

FAQs

What is the main idea of the proposed detection method?

The method computes new features from biomarker data, normalizes them using Gaussian distribution membership, and arranges them as inputs for a ResNet50 neural network to classify breast cancer.

How are the features normalized in the approach?

Original feature values are transformed into normalized values based on their membership in a Gaussian distribution built from healthy and breast-cancer samples for each feature.

Which datasets are used to test the approach and what results are reported?

The Coimbra and Wisconsin datasets are used, with reported accuracy of 100% on the Coimbra dataset and 100% on the Wisconsin dataset.

Sep 21, 2026

Uploaded by Miles