Breast imaging is transforming as researchers explore new methods to interpret mammograms more efficiently and accurately.
Given the demands of a high-volume, high-stakes field, innovations that can streamline workflows while improving diagnostic accuracy are crucial. Two recent studies provide insights into how structured ordering methods for reading mammograms could improve breast imaging practices.
By focusing on ordering mammogram cases based on specific characteristics, such as breast density or visual similarity, these studies explore whether changing the interpretation sequence could reduce radiologists’ cognitive load, enhance diagnostic performance, and ultimately benefit patients and practitioners.
The Challenge in Mammogram Interpretation
Mammogram interpretation requires radiologists to scrutinize each image, identifying small abnormalities that could indicate early-stage cancer. This work is essential but demanding: in high-volume practices, radiologists read through hundreds of images daily, which can lead to cognitive fatigue, increased stress, and risk of burnout. Adding to these challenges, radiologists must balance accuracy with efficiency, a difficult feat given the shortage of breast imaging specialists in many regions.
The studies examined here are part of a growing body of research exploring how structured workflows could provide much-needed relief. These methods focus on optimizing the sequence in which mammograms are interpreted to improve accuracy, reduce reading times, and mitigate radiologist fatigue.
Study Insights: Ordering by Volumetric Breast Density (VBD) vs. AI-Driven Ordering of Images According to Visual Similarity
The first study by Gommers and Verboom et al. [1] evaluated an ordering approach based on Volumetric Breast Density (VBD). This method structures mammograms to increase breast density, allowing radiologists to start with low-density (easier-to-interpret) cases and gradually transition to high-density (more complex) cases.
In their study, 13 experienced radiologists reviewed 150 mammograms in three different sequences: VBD ordering, random ordering (the current standard), and AI-driven similarity-based ordering, which grouped mammograms based on visual similarity.
The results demonstrated a significant advantage for the VBD approach. Radiologists showed a higher diagnostic performance with VBD ordering, evidenced by an increase in the area under the receiver operating characteristic (ROC) curve (AUC) from 0.92 for random ordering to 0.93 with VBD ordering (P = .009). Additionally, the VBD approach led to faster reading times (24.3 seconds vs. 27.9 seconds for random ordering), fewer eye fixations on images, and reduced time spent fixating on malignant regions. These metrics suggest that VBD ordering reduces radiologists’ cognitive load, allowing them to detect abnormalities faster and more accurately.
In contrast, AI-driven similarity ordering (SSL = self-supervised learning), which grouped images based on visual appearance, did not improve performance and resulted in longer reading times than random ordering. This suggests that while AI holds promise in radiologic applications, this particular AI strategy did not align well with the radiologists’ perceptual and diagnostic processes.
However, researchers believe AI may still have a role in mammogram ordering, especially as algorithms are refined to account for radiologist fatigue and individual strengths.
Why Visual Adaptation Matters
The VBD-based ordering method leverages a principle known as visual adaptation. In visual adaptation, a person’s visual perception adjusts gradually to changes in their environment—in this case, transitioning from low-density, more straightforward images to high-density, more complex ones.
As radiologists progress through the cases, they adjust to the increasing complexity, allowing for more efficient processing of dense tissue, which can often obscure early signs of cancer. This sequencing minimizes the cognitive fatigue that can arise from an unpredictable order of cases, allowing radiologists to maintain high performance throughout their reading sessions.
Patient-Centric Benefits
For patients, these improvements in workflow and diagnostic accuracy translate into tangible benefits:
- Improved Diagnostic Accuracy and Early Detection: Enhanced performance in identifying subtle abnormalities in dense breast tissue means a greater chance of detecting cancers early, particularly in women with dense breasts who may otherwise face a higher risk of missed diagnoses.
- Faster Results and Reduced Anxiety: Shorter reading times allow radiologists to process mammograms more quickly, reducing the wait time for patients to receive results. Quicker turnaround times lessen the anxiety that often accompanies the wait for potentially life-altering news, particularly for patients awaiting follow-up imaging or biopsies.
- Reduced Unnecessary Recalls: More accurate readings help reduce false-positive results, which means fewer unnecessary recalls for additional imaging. These follow-up appointments can be stressful, costly, and inconvenient for patients, so reducing their frequency can significantly improve the patient experience.
- Greater Access to Screening: The efficiency gains associated with VBD ordering allow practices to accommodate more screenings within the same time frame. This increased capacity benefits all patients, particularly those in high-demand areas, by improving access to timely screenings and preventive care.
Benefits to Radiologists and Practices
The advantages of a structured workflow extend beyond patient outcomes, impacting the day-to-day operations of imaging practices and the well-being of radiologists:
- Mitigating Burnout: The structured sequencing associated with VBD ordering helps reduce cognitive load and may lower the risk of burnout, a significant concern in radiology today. By starting with more straightforward cases, radiologists can ease into more complex cases, creating a workflow that aligns with natural cognitive rhythms rather than imposing unpredictable demands.
- Increased Efficiency and Throughput: Shortened reading times mean more cases can be processed daily. For high-volume practices, this can reduce backlogs, improve patient flow, and decrease the likelihood of diagnosis delays.
- Cost Savings and Resource Optimization: Improved efficiency translates into cost savings, as practices can reduce overtime or require additional staff to manage large caseloads. Optimized workflows enable serving more patients without adding strain to existing resources.
- Improved Diagnostic Consistency: By creating a structured reading order, VBD ordering reduces reading variability, potentially leading to more consistent diagnostic outcomes. This consistency benefits practices by supporting quality control efforts and helping radiologists meet key performance metrics in cancer detection and recall rates.
Future Directions: AI and Personalized Reading Orders
Although the current AI-driven approach did not yield better outcomes than VBD ordering, AI still holds potential for future improvements. With further development, AI algorithms could be designed to adapt case order based on real-time data about a radiologist’s performance, adjusting for fatigue levels or specific diagnostic strengths and weaknesses. Personalized reading orders created by AI could one day customize workflows to each radiologist, potentially maximizing efficiency and accuracy in a way that VBD ordering alone might not achieve.
Conclusion
While the authors acknowledge that their study is on the cutting edge of research in this area, a VBD-based ordering system represents a promising advancement in optimizing breast cancer screening workflows.
By aligning case sequences with radiologists’ perceptual strengths, VBD ordering supports a more accurate, efficient, and patient-centered approach to care. This combination of improved diagnostic accuracy, reduced reading times, and decreased cognitive load has the potential to reshape breast imaging practices, directly benefiting both patients and healthcare providers.
As the healthcare field embraces new technologies such as AI and integrates perceptual science principles like visual adaptation, these findings lay the groundwork for further innovations. Future studies that refine these methods and explore applications in other imaging modalities could deepen our understanding of how best to support radiologists in high-stakes, high-volume environments.
Ultimately, advances like these help us envision a future where breast cancer screening is more efficient, accessible, reliable, and supportive of the patients it serves.
References
Gommers, Jessie J. J., et al. ” Enhancing Radiologist Reading Performance by Ordering Screening Mammograms Based on Characteristics That Promote Visual Adaptation.” Radiology, vol. 313, no. 1, 2024, pp. e240237, https://doi.org/10.1148/radiol.240237.