Enhancing Histopathological Image Classification of Invasive Ductal Carcinoma Using Hybrid Harmonization Techniques

Sep 24, 2026

Uploaded by Mary Man

Robust histopathological image classification for invasive ductal carcinoma and benign tumors addresses variability within and between centers. The study builds a feature-engineering pipeline with feature extraction and multiple harmonization steps to correct intra- and inter-center batch effects caused by acquisition differences and heterogeneous patient characteristics. Performance is evaluated on BreaKHIS and IDCDB using recall, precision, and accuracy. Results exceed recent feature-engineering baselines with up to 90–95% accuracy, highlighting harmonization benefits for patch learning and testing across databases.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages16

Reading Time40 min

AccessFree to download

Background

Batch effects and data pooling across centers

Challenges in histopathological harmonization

Proposed approach

Feature extraction pipeline

Intra-database variability and intra-slide harmonization

Inter-database variability and cross-database harmonization

Evaluation

Datasets: BreaKHIS and IDCDB

Metrics and performance comparison

Impact and applications

[www. nature.com/scientificreports](www. nature.com/scientificreports) OPEN Enhancing histopathological image classification of invasive ductal carcinoma using hybrid harmonization techniques Nassib Abdallah1,2*, Jean‑Marie Marion3, Clovis Tauber4, Thomas Carlier5, Mathieu Hatt1 & Pierre Chauvet3 This study aims to develop a robust pipeline for classifying invasive ductal carcinomas and benign tumors in histopathological images, addressing variability within and between centers. We specifically tackle the challenge of detecting atypical data and variability between common clusters within the same database. Our feature engineering‑based pipeline comprises a feature extraction step, followed by multiple harmonization techniques to rectify intra‑ and inter‑center batch effects resulting from image acquisition variability and diverse patient clinical characteristics. These harmonization steps facilitate the construction of more robust and efficient models. We assess the proposed pipeline’s performance on two public breast cancer databases, BreaKHIS and IDCDB, utilizing recall, precision, and accuracy metrics. Our pipeline outperforms recent models, achieving 90‑95% accuracy in classifying benign and malignant tumors. We demonstrate the advantage of harmonization for classifying patches from different databases. Our top model scored 94.7% for IDCDB and 95.2% for BreaKHis, surpassing existing feature engineering‑based models (92.1% for IDCDB and 87.7% for BreaKHIS) and attaining comparable performance to deep learning models. The proposed feature‑ engineering‑based pipeline effectively classifies malignant and benign tumors while addressing variability within and between centers through the incorporation of various harmonization techniques. Our findings reveal that harmonizing variabilities between patches from different batches directly impacts the learning and testing performance of classification models. This pipeline has the potential to enhance breast cancer diagnosis and treatment and may be applicable to other diseases. One of the major challenges in biomedical research lies in the necessity to have substantial volumes of patient data to train classification models effectively. However, in the vast majority of biomedical applications, such extensive datasets are not readily available. Consequently, we often resort to pooling patient data from multiple acquisition centers. This practice introduces what is commonly known as the “batch effect,” an artifact attributed to differences in acquisition hardware or protocols. Such batch effects hinder the generalizability of our models. Data harmonization is becoming an increasingly crucial issue in biomedical research. This is done by estimating the batch effect between different centers and minimizing its impact, thereby enhancing the generalizability of the models. This technique has been extensively applied in medical imaging, in FDG PET/CT imaging, as seen in works like1 and2 and in MRI imaging3. However, harmonization of features in histopathological slices is less widespread. Harmonization constitutes a critical, yet intricate, facet of histopathological image classification. Specifically, two principal types of variability serve as obstacles to the robust performance of machine learning algorithms: intra-database and inter-database. Intra-database variability arises from inconsistencies present within a single data collection center, often taking the form of variations in staining or fluctuations in quality across patches within an individual histopathological slide. Figure 5 illustrates examples of patches from a histopathological slide following unsupervised clustering. One cluster comprises images of border regions, while the other encapsulates images of central regions, thereby underscoring the need for intra-slide harmonization to mitigate such variabilities. Additionally, this intra-slide 1LaTIM, INSERM, Université de Bretagne-Occidentale, Brest, France. 2

FAQs

What problem does the proposed pipeline address in histopathological classification?

It targets variability within and between acquisition centers that creates batch effects and degrades generalizability. It also focuses on detecting atypical data and differences across clusters.

How does the pipeline reduce intra- and inter-center batch effects?

It uses a feature-engineering workflow with feature extraction followed by multiple harmonization techniques. These steps rectify batch effects from image acquisition variability and diverse patient characteristics.

On which datasets is the method evaluated and how well does it perform?

The pipeline is assessed on two public breast cancer databases, BreaKHIS and IDCDB, using recall, precision, and accuracy. It achieves top performance of 94.7% on IDCDB and 95.2% on BreaKHis.

Sep 24, 2026

Uploaded by Mary Man