AI Boosts Cancer Detection While Cutting Radiologists’ Workload

In the complex landscape of modern oncology, breast cancer screening faces a daunting paradox: while imaging technology advances, the pressure on radiologists reaches critical levels due to a shortage of specialists and an increasing workload. However, the results of the recently published GEMINI study ( Grampian Evaluation of Mia in an Innovative National Breast Screening Initiative ), disseminated by the journal Nature Cancer, promise to be a game-changer in how we detect this disease.

This prospective study, conducted in the Grampian region (UK), has evaluated 10,889 women to demonstrate that the intelligent integration of Artificial Intelligence (AI) is not only safe, but can increase the cancer detection rate by 10.4% while reducing the workload of physicians by up to 31% .

For decades, the standard in European countries has been dual human review : two radiologists analyze each mammogram independently. If there is disagreement, a third expert arbitrates. Although effective, this system is intensive in human resources , an increasingly scarce commodity in modern healthcare systems.

The GEMINI study, led by researcher Clarisse Florence de Vries and a multidisciplinary team from NHS Grampian and the University of Aberdeen, didn’t just test whether AI works, but explored 17 different ways to integrate it into the clinical workflow. The goal was clear: to find the perfect balance between diagnostic accuracy and operational efficiency.

Mia v.3 and deep learning

The tool featured in the study was Mia v.3, an AI system based on deep learning designed to analyze mammograms and issue a recommendation of “recall”, to schedule the patient for further testing, or “no recall”.

What makes GEMINI unique is that it used AI in two complementary functions. The first was as an “additional reading ,” identifying suspicious cases that the two human readers had missed, and the second was as a triage tool to decide which cases required a second human reading and which could be validated more quickly.

The Mia v.3 system is not a conventional algorithm; it is a deep learning tool trained for over ten years using data from screening programs in various countries and from multiple hardware manufacturers. Based on each woman’s digital mammograms, the AI ​​generates a continuous malignancy prediction value ranging from 0 to 1, recommending further examination whenever this value exceeds a predefined decision threshold.

A key aspect of the study is that no prior data was used to train or calibrate the model, ensuring that the results reflect true generalizability and a completely technology- independent evaluation in a novel clinical setting.

The versatility of this tool lies in its operating points (OPs) , which are thresholds predefined by the provider to balance sensitivity and specificity according to the objectives of each healthcare service. During the GEMINI study, AI was used in vivo under OP2, a configuration designed to maximize the detection of potential cancers, but simulations were also evaluated using OP1 to seek greater specificity and time savings.

This flexibility allowed for the modeling of scenarios where AI not only acts as an additional “safety net”, but can even replace the second human reader in cases where the algorithm and the first radiologist agree on the absence of suspicious signs.

The 11 cases that would have gone unnoticed

One of the most striking findings of the study is the detection of 11 additional cancers thanks to AI support. In these cases, routine human double-checking had not recommended recalling the patient, but the AI ​​flagged the images as suspicious. Specifically, it identified 1,345 cases, and 55 women were referred for further testing, of which 11 tested positive.

Following further human review of these AI-tagged cases, tumors were confirmed that would otherwise have had to wait at least three years—the standard screening interval in the UK—to be detected, with the consequent risk of disease progression. Of these 11 cancers, 7 were invasive, including grade 2 and 3 tumors.

The GEMINI study stands out for its pragmatism. The researchers understand that not all hospitals have the same needs. Some may prioritize maximum screening, while others, overwhelmed, may need to alleviate the workload of their specialists.

To do this, they evaluated different operating points, which are decision thresholds that adjust the sensitivity and specificity of the AI.

The primary workflow combined triage of negative cases with additional AI reading. The result was an increase of 1 cancer detected per 1,000 women screened, without increasing the rate of unnecessary appointments (false positives) and saving almost a third of the human reading work.
Maximum work savings. Other simulated models showed that the workload could be reduced by up to 44% if AI acts as a second reader in cases where it overlaps with the first human radiologist.

Speed ​​and human control: AI as support, not as a replacement

One of the recurring concerns about AI is whether it will slow down the clinical process or whether doctors will blindly trust it. GEMINI sheds light on this: 63% of the additional reviews suggested by AI were resolved by radiologists in less than 30 seconds.

The experts demonstrated remarkable critical ability, quickly ruling out AI markers that corresponded to previous surgical scars or benign calcifications. This suggests that AI acts as a “safety net” that the physician can efficiently manage without losing ultimate control of the decision.

The GEMINI findings add to a growing body of evidence in Europe. Previous studies, such as the MASAI trial in Sweden, reported similar results, with an increase in cancer detection of 1 per 1,000 women and a 44.3% reduction in workload. In Denmark, the routine use of AI has already led to a 33.5% reduction in workload.

GEMINI’s contribution lies in its forward-looking approach and the evaluation of multiple implementation strategies in a “real-world” environment, making it easier for other screening programs to choose the configuration that best suits their local resources.

Challenges and limitations on the horizon

Despite the success, the study also points to areas for improvement. The AI ​​system excluded 10.4% of mammograms for not meeting its technical criteria, meaning that there is still a proportion of patients who rely exclusively on the human eye.

Furthermore, the researchers caution that changes to mammography software or hardware can affect AI performance, necessitating constant monitoring and rigorous quality assurance processes . A three-year follow-up study is also needed to analyze “interval cancers”—those that appear between two screening rounds—and determine if AI helps reduce them.

By demonstrating that AI can be configured to meet diverse clinical and operational needs, it opens the door to personalized screening . The Mia v.3 technology and the GEMINI model demonstrate that when human and artificial intelligence collaborate, the result is a more robust, efficient healthcare system, and above all, one capable of detecting the invisible.

__

(Featured image by Sasun Bughdaryan via Unsplash)

DISCLAIMER: This article was written by a third party contributor and does not reflect the opinion of Born2Invest, its management, staff or its associates. Please review our disclaimer for more information.

This article may include forward-looking statements. These forward-looking statements generally are identified by the words “believe,” “project,” “estimate,” “become,” “plan,” “will,” and similar expressions. These forward-looking statements involve known and unknown risks as well as uncertainties, including those discussed in the following cautionary statements and elsewhere in this article and on this site. Although the Company may believe that its expectations are based on reasonable assumptions, the actual results that the Company may achieve may differ materially from any forward-looking statements, which reflect the opinions of the management of the Company only as of the date hereof. Additionally, please make sure to read these important disclosures.

First published in GACETA MEDICA. A third-party contributor translated and adapted the article from the original. In case of discrepancy, the original will prevail.

Although we made reasonable efforts to provide accurate translations, some parts may be incorrect. Born2Invest assumes no responsibility for errors, omissions or ambiguities in the translations provided on this website. Any person or entity relying on translated content does so at their own risk. Born2Invest is not responsible for losses caused by such reliance on the accuracy or reliability of translated information. If you wish to report an error or inaccuracy in the translation, we encourage you to contact us.