A Green Solution for Breast Region Segmentation Using Deep Active Learning

TL;DR

This work addresses the annotation burden and environmental impact of training breast region segmentation models on MR images by introducing a BAG-based active learning framework. Using FCN-ResNet50, four sampling strategies are compared across four data proportions (10–40%), evaluated with 5-fold cross-validation and metrics including $Dice$, $IoU$, $Precision$, $Recall$, and $HD$. The key finding is that the Nearest Point sampling strategy generally yields strong segmentation performance with a lower carbon footprint, particularly around 30% of the data, suggesting a practical, eco-friendly data-efficient approach for BRS. The study demonstrates how breast anatomy geometry-informed sampling can reduce labeled data needs while maintaining accuracy, offering significant implications for sustainable AI in medical imaging.

Abstract

Purpose: Annotation of medical breast images is an essential step toward better diagnostic but a time consuming task. This research aims to focus on different selecting sample strategies within deep active learning on Breast Region Segmentation (BRS) to lessen computational cost of training and effective use of resources. Methods: The Stavanger breast MRI dataset containing 59 patients was used in this study, with FCN-ResNet50 adopted as a sustainable deep learning (DL) model. A novel sample selection approach based on Breast Anatomy Geometry (BAG) analysis was introduced to group data with similar informative features for DL. Patient positioning and Breast Size were considered the key selection criteria in this process. Four selection strategies including Random Selection, Nearest Point, Breast Size, and a hybrid of all three strategies were evaluated using an active learning framework. Four training data proportions of 10%, 20%, 30%, and 40% were used for model training, with the remaining data reserved for testing. Model performance was assessed using Dice score, Intersection over Union, precision, and recall, along with 5-fold cross-validation to enhance generalizability. Results: Increasing the training data proportion from 10% to 40% improved segmentation performance for nearly all strategies, except for Random Selection. The Nearest Point strategy consistently achieved the lowest carbon footprint at 30% and 40% data proportions. Overall, combining the Nearest Point strategy with 30% of the training data provided the best balance between segmentation performance, efficiency, and environmental sustainability. Keywords: Deep Active Learning, Breast Region Segmentation, Human-center analysis

Figures (8)

• Figure 1: Schematic illustration of the proposed active learning cycle
• Figure 2: Illustration of different breast anatomy geometries including Nearest Point and Breast Size by using breast region overlay map
• Figure 3: Block diagram of the FCN-ResNet50 architecture showing specifications of each layer
• Figure 4: Breast Size and Nearest Point histogram along all patients
• Figure 5: Segmentation results across all strategies for different slices for test dataset by using model trained with 10 and 20 percent of whole dataset