Segmentation, Classification and Interpretation of Breast Cancer Medical Images using Human-in-the-Loop Machine Learning - doctor-in-the-loop approach

Aug 3, 2026

Uploaded by Aria

This paper explores Human-in-the-Loop (HITL) strategies to train machine learning models in medical imaging, proposing a “doctor-in-the-loop” approach to use expert knowledge for large and complex datasets. The work combines genomic data with Whole Slide Imaging (WSI) analysis for breast cancer, addressing segmentation of histopathology images, classification by genomic subtype, and interpretation of model outputs. A pathologist improved segmentation quality and enhanced model explainability, while classification remained suboptimal, indicating that complex domains can limit HITL effectiveness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages32

Reading Time1 h 21 min

AccessFree to download

Introduction

Breast cancer heterogeneity and clinical diagnosis

Genomic vs immunohistochemical classification

Motivation for histopathology image analysis

Human-in-the-loop collaboration

Doctor-in-the-loop process

Segmentation of histological slides

arXiv :2403 .20112v1 [ cs .CV] 29 Mar 2024 Segmentation, Classification and Interpretation of Breast Cancer Medical Images using Human-in-the-Loop Machine Learning David V´azquez-Lema 1 , Eduardo Mosqueira-Rey 1*, Elena Hern´andez-Pereira 1 , Carlos Fern´andez-Lozano 1 , Fernando Seara-Romera 1 , Jorge Pombo-Otero2 1* Department of Computer Science and Information Technologies, University of Coru˜na (CITIC), R´ua Maestranza, 9, La Coru˜na, 15001, Galicia, Spain. 2* Servicio de Anatom´ıa Patol´ogica, Complejo Hospitalario Universitario de A Coru˜na (CHUAC), As Xubias, 84, A Coru˜na, 15006, Spain. *Corresponding author(s) . E-mail(s): [eduardo@udc.es](eduardo@udc.es) ; Contributing authors: [david.vazquez7@udc.es](david.vazquez7@udc.es) ; [elena.hernandez@udc.es](elena.hernandez@udc.es) ; [carlos.fernandez@udc.es](carlos.fernandez@udc.es) ; [f.searar@udc.es](f.searar@udc.es) ; [jorge.pombo.otero@sergas.es](jorge.pombo.otero@sergas.es) ; Abstract This paper explores the application of Human-in-the-Loop (HITL) strategies in training machine learning models in the medical domain. In this case a ‘doctor-inthe-loop’ approach is proposed to leverage human expertise in dealing with large and complex data. Specifically, the paper deals with the integration of genomic data and Whole Slide Imaging (WSI) analysis of breast cancer. Three different tasks were developed: segmentation of histopathological images, classification of this images regarding the genomic subtype of the cancer and, finally, interpretation of the machine learning results. The involvement of a pathologist helped us to develop a better segmentation model and to enhance the explainatory capabilities of the models, but the classification results were suboptimal, highlighting the limitations of this approach: despite involving human experts, complex domains can still pose challenges, and a HITL approach may not always be effective. Keywords: Human-in-the-loop, Breast cancer, Segmentation, Classification, Interpretation 1 1 Introduction Cancer is a highly heterogeneous disease and a major contributor to global mortality, responsible for about 1 in every 6 deaths [1] . Cancer cells are often undifferentiated, complicating their characterization, and exhibit a high capacity for proliferation/migration within the tumor microenvironment. Different subtypes are often identified for each specific type of cancer, with varying prognoses, survival times, and responses to treatments. For women, breast cancer (BC) has surpassed lung cancer as the most diagnosed cancer worldwide accounting for 32% of cases [1, 2], and 23% of mortality. Overall, breast cancer, lung cancer, and colorectal cancer account for 51% of all new diagnoses annually for women. The incidence rates of female breast cancer have experienced a gradual rise of approximately 0.6% annually since the mid-2000s, primarily attributed to the increased detection of localized-stage and hormone receptor-positive cases [3] . Currently, the diagnosis of BC in daily clinical practice is made by immunohistochemical analysis of cancerous breast tissue removed during surgery, studying the presence or absence of estrogen receptor (ER), progesterone receptor (PR), HER2 membrane protein, and the Ki-67 proliferation index. In those healthcare centers with the capacity for massive sequencing, the most common approach is the study of the tumor cell transcriptome based on the expression of 50-gene signature [4] in paraffin-embedded tumor tissue, also called intrinsic subtypes (IS) . There is no perfect correlation between immunohistochemical and genomic classification [5], with the latter being more accurate but much more expensive and less common in daily clinical practice. As mentioned earlier, in daily clinical practice in hospitals, and despite the rapid advances in genomic medicine, pathologists still carry out patient stratification through manual review of histopathology slides with a light microscope.

FAQs

What human-in-the-loop strategy is proposed for this breast cancer study?

The paper proposes a “doctor-in-the-loop” approach where a pathologist is actively involved to incorporate expert knowledge during model development.

Which tasks are implemented using genomic data and Whole Slide Imaging?

Three tasks are developed: segmentation of histopathological images, classification according to genomic subtype, and interpretation of the machine learning results.

How did involving a pathologist affect the model performance?

Pathologist involvement helped produce a better segmentation model and improved the explainatory capabilities of the models, while classification results were suboptimal.

Aug 3, 2026

Uploaded by Aria