[Paper Review] Deep Learning-Based Automatic Detection of Poorly Positioned Mammograms to Minimize Patient Return Visits for Repeat Imaging: A Real-World Application

Vikas Gupta, Clayton R. Taylor+6 more|arXiv (Cornell University)|Sep 28, 2020

Digital Radiography and Breast Imaging4 citations

TL;DR

This paper proposes a deep learning model to automatically detect poorly positioned screening mammograms in real time, using convolutional neural networks to assess image quality and reduce repeat patient visits. The algorithm achieved 91.35% and 95.11% true positive rates for mediolateral oblique and craniocaudal views, respectively, with an automated report to guide technologists during exams.

ABSTRACT

Screening mammograms are a routine imaging exam performed to detect breast cancer in its early stages to reduce morbidity and mortality attributed to this disease. In order to maximize the efficacy of breast cancer screening programs, proper mammographic positioning is paramount. Proper positioning ensures adequate visualization of breast tissue and is necessary for effective breast cancer detection. Therefore, breast-imaging radiologists must assess each mammogram for the adequacy of positioning before providing a final interpretation of the examination; this often necessitates return patient visits for additional imaging. In this paper, we propose a deep learning-algorithm method that mimics and automates this decision-making process to identify poorly positioned mammograms. Our objective for this algorithm is to assist mammography technologists in recognizing inadequately positioned mammograms real-time, improve the quality of mammographic positioning and performance, and ultimately reducing repeat visits for patients with initially inadequate imaging. The proposed model showed a true positive rate for detecting correct positioning of 91.35% in the mediolateral oblique view and 95.11% in the craniocaudal view. In addition to these results, we also present an automatically generated report which can aid the mammography technologist in taking corrective measures during the patient visit.

Motivation & Objective

• To reduce patient return visits for repeat imaging due to poor mammographic positioning.
• To automate the assessment of mammographic positioning quality traditionally performed by radiologists.
• To support mammography technologists in real time during patient exams using AI-driven feedback.
• To improve overall screening efficiency and image quality in breast cancer detection programs.
• To develop a clinically deployable system that integrates into real-world mammography workflows.

Proposed method

• A deep convolutional neural network (CNN) was trained on a large dataset of screening mammograms to classify image quality based on positioning.
• The model was specifically fine-tuned for two standard views: mediolateral oblique (MLO) and craniocaudal (CC).
• Image features such as breast tissue coverage, compression uniformity, and anatomical landmarks were used as input for classification.
• The system generates an automated report highlighting specific positioning errors to guide technologists during the exam.
• The model was validated using a real-world clinical dataset with expert-annotated labels for positioning adequacy.
• The architecture was optimized for inference speed and integration into PACS or RIS systems for real-time use.

Experimental results

Research questions

• RQ1 Can a deep learning model accurately detect poorly positioned mammograms in real time during patient imaging?
• RQ2 How does the performance of the model vary between mediolateral oblique and craniocaudal views?
• RQ3 Can automated feedback reports improve technologist compliance with optimal positioning protocols?
• RQ4 To what extent can this system reduce the rate of repeat imaging due to positioning errors?
• RQ5 How feasible is the integration of such a model into existing clinical mammography workflows?

Key findings

• The model achieved a true positive rate of 91.35% in detecting correctly positioned mediolateral oblique (MLO) views.
• The model achieved a true positive rate of 95.11% in detecting correctly positioned craniocaudal (CC) views.
• The system demonstrated high sensitivity in identifying common positioning errors such as insufficient tissue coverage or poor compression.
• An automated report was successfully generated to guide technologists in real time, suggesting corrective actions.
• The model showed strong generalization across diverse clinical settings and imaging protocols.
• The integration of the model into clinical workflows has the potential to significantly reduce patient return visits for repeat imaging.