MASAI Trial: AI Can Help Radiologists Identify Breast Cancer 

The Mammography Screening with Artificial Intelligence (MASAI) trial suggests that AI could improve the accuracy of breast cancer screening by up to 29%, while massively reducing clinical workload. 

Breast Cancer Screening Today 

Despite ongoing work in research and public education, breast cancer is a prevalent and deadly disease, with approximately 150 new diagnoses per day in the UK alone and a 10-year survival rate of roughly 76%. Early diagnosis maximizes the chance of successful treatment, and the introduction of national screening programmes across Europe has been instrumental in identifying signs of tumors before they become symptomatic.  

Though the availability, frequency, and criteria for mammography screening may vary between each country, it is generally recommended that each scan is reviewed by two radiologists, with further review if required. Due to the specialized skills required to accurately interpret mammography scans, breast radiologists are in short supply, and interpreting the results of national screening programs is a massive undertaking.  

What was the MASAI trial? 

Conducted alongside the Swedish National Screening Programme, the MASAI trial sought to investigate whether the introduction of AI could both improve the accuracy of mammography results and lighten the workload of breast radiologists.  

The study included over 105,000 patients aged 40 to 74 across Sweden, who were subsequently divided into:

• The control group, whose results were analysed by standard double-reading.
• The intervention group, where the radiologists were assisted by AI.  

Scans from the intervention group were triaged by ScreenPoint Medical’s deep learning software Transpara AI, which flagged suspicious areas on the image and assigned each patient a risk-based score from one to ten (with ten being high risk). Scans with a lower score were referred for review by one radiologist, whereas riskier scans were reviewed by at least two, and particularly worrying scans were highlighted for prioritised viewing. 

Results 

Remarkably, the introduction of Transpara appeared to increase the number of cancers identified by roughly 29% (6.4 per thousand compared with 5 per thousand in the control group). Beyond identifying a higher total number of invasive cancers (270 versus 217 | AI-led versus human-led screening), radiologists using AI also identified 46 more invasive cancers that were lymph node-negative and 58 more that were smaller than 2 cm, something that is easily missed without the AI.  

This could potentially hint that the use of AI helped radiologists pick up on subtle signs of invasive tumours in their earliest stages. It was also noted that despite the higher number of malignancies identified in the intervention group, they did not have a significantly higher number of false positive results. 

Perhaps most exciting, doctors using Transpara identified more patients with cancer subtypes known to be particularly aggressive or difficult to treat (16 triple negative cancers were identified in the intervention group (AI-led) versus 6 in the control group, and 17 HER2 positive patients in the AI-led group versus 13 with standard screening). 

Finally, the study noted that the control group required a total of 109,692 “screen readings” from radiologists, compared with 61,248 in the intervention group. Based on this, it was estimated that integration of AI cut down the radiology workload by approximately 44%.  

Public Perception 

Since being published in The Lancet Digital Health, the MASAI trial has been hailed as a landmark study by Oliver Elemento, Director of the Englander Institute for Precision Medicine, and Associate Director of the Institute for Computational Biomedicine, who argues that:

“This trial is a landmark in demonstrating that AI in medicine can and should be tested under the same rigorous standards as new drugs and medical devices.”

This trial demonstrates a practical application of AI to help solve a real-world challenge. Though it has drawn praise from online media platforms such as FemTech World and Oncodaily, the study has ignited discussion on social media with several arguing that the trial does not disclose their false positive rates and that the use of AI in such large-scale diagnostics relies on several existing factors. 

For example, there needs to be a standardised and established screening programme akin to the one used for this study, as well as a centralised data system to store and access the results, features that may not be feasible for many populations.  

Regardless, whilst more research will be needed to support the conclusions drawn from the MASAI trial, the methodology shows how AI could theoretically be safely implemented into current clinical practice. As stated by lead researcher Kristina Lång:  

“Our findings indicate that AI-supported screening can significantly enhance the early detection of clinically relevant breast cancers while reducing the workload for radiologists. This has the potential to improve patient outcomes and optimise the use of healthcare resources.” 

Other AI Screening Tools 

The Swedish MASAI trial is not alone in this approach—of using AI to detect breast cancer. Many other countries and research institutes are developing AI tools for the same purpose: 

1. Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory and Massachusetts General Hospital developed a deep-learning model that can predict breast cancer occurrence from mammograms.  
2. Researchers at Duke University have developed AsymMirai, an interpretable deep-learning model that focuses on breast tissue asymmetry. By comparing tissue patterns between the left and right breasts, AsymMirai provides a streamlined yet accurate method for predicting breast cancer risk. 
3. The National Health Service in the UK is testing an AI system called Mammography Intelligent Assessment (MIA). MIA acts as a second reader, scrutinizing mammograms for subtle signs of cancer that radiologists might miss, thereby offering an added layer of protection. 

How Does Early Detection Help? 

Early detection of any cancer is crucial for survival rate. As more time progresses, cancer does too, and by the time it reaches stage 2/3, sometimes it is too advanced and too metastasized to be treated. Early detection means that cancer can be treated whilst still in its early stages, reducing the chances of metastasis.  

AI is a tool that is needed for national screening programmes, as mentioned, there aren’t enough radiologists for this massive undertaking. The use of AI also makes it possible for countries to do these programmes, as it not only cuts costs tremendously but also cuts the amount of time needed to screen each patient.  

The MASAI trial, along with other similar AI-driven screening initiatives, highlights the transformative potential of AI in breast cancer detection. By improving accuracy, reducing radiologists’ workload, and enabling earlier identification of aggressive cancer subtypes, AI has the capability to enhance patient outcomes and patient healthcare resources.