Could Genius AI Detection technology from Hologic be the next breakthrough in early detection of breast cancers? By locating lesions that are likely to represent breast cancer in each slice of a digital breast tomosynthesis (DBT) image, the deep-learning algorithm is designed to aid radiologists in confident breast cancer detection following mammography.
Dr. Nisha Sharma, consultant breast radiologist and director of the breast screening programme for Leeds and Wakefield, Leeds Teaching Hospitals NHS Trust, U.K., has been evaluating Genius AI Detection v2.0 in a symptomatic breast setting since August 2024 and shares her positive experience with the innovative technology.
"In the symptomatic service, we are very busy -- an AI tool that can help read the multiple images produced by tomosynthesis and flag priority cases is invaluable," she commented.
Could Genius AI Detection technology from Hologic be the next breakthrough in early detection of breast cancers? By locating lesions that are likely to represent breast cancer in each slice of a digital breast tomosynthesis (DBT) image, the deep-learning algorithm is designed to aid radiologists in confident breast cancer detection following mammography.
Dr. Nisha Sharma, consultant breast radiologist and director of the breast screening programme for Leeds and Wakefield, Leeds Teaching Hospitals NHS Trust, U.K., has been evaluating Genius AI Detection v2.0 in a symptomatic breast setting since August 2024 and shares her positive experience with the innovative technology.
"In the symptomatic service, we are very busy -- an AI tool that can help read the multiple images produced by tomosynthesis and flag priority cases is invaluable," she commented.
Dr. Nisha Sharma.
“We are excited that DBT has been embedded within our symptomatic clinical workflow because we know it is a really good imaging tool, both in terms of increasing the reader’s confidence and also in determining whether any abnormality is likely to be benign or malignant,” Sharma explained.
However, she noted that a key challenge with tomosynthesis is the time taken to read multiple imaging slices. “Here, Genius AI Detection is highly beneficial. It can be additionally applied for women who have tomosynthesis imaging to improve the cancer detection rate and accelerate workflow,” she commented.
AI technology has been used in radiology for many years, with mixed results. The "ImageChecker Computer-Aided Detection (CAD)" system was first introduced in 2006 and used traditional techniques to teach software to look for patterns in images identified as good proxies for cancerous lesions. However, diagnostic accuracy was not always improved with CAD algorithms due to low specificity and high false-positive rates.
Unlike CAD algorithms, deep learning-based AI algorithms do not rely on human-derived or manually engineered imaging features. Deep learning-based AI algorithms use convolutional neural networks to extract high-level imaging features from large amounts of raw data, and the model itself learns the features and associations to identify malignancy. Indeed, a deep learning-based AI algorithm on DBT images was found to significantly outperform a traditional CAD algorithm applied to synthetic 2D mammography.1
Safety net and prioritization tool
Sharma said the Genius AI Detection technology assists with clinical decision-making by providing a valuable second pair of eyes. "As radiologists, we make decisions based on initial investigations, but AI-guided tools now provide a second opinion. If AI concurs with our findings that is great, and if it flags something additional, we can either dismiss it (based on results of a previous mammogram or imaging) or investigate further, so it is a valuable safety net."
She emphasized that the AI algorithm aids the decision-making process but that clinical expertise and knowledge gained over many years are also extremely valuable, and she would never rely solely on AI if she had clinical concerns.
According to Sharma, AI implementation is timely as the breast radiologist workforce is declining. Currently, they double report (i.e., results are checked by two radiologists), but in the future, this may be difficult to deliver, and double reporting may only be possible with the support of an AI algorithm as a "reader assist."
Sharma also describes Genius AI Detection as a prioritization tool; by helping to determine cases that are more or less likely to require workup, the team are more control of their workflow and able to prioritize certain patients the same day. An AI-based approach has demonstrated a high cancer detection rate, but the algorithm does generate some false positives. She acknowledges that this is not always inappropriate: "The AI algorithm may flag cases that are benign, but not obviously benign without further workup, so these findings are useful. False positives can usually be readily dismissed and accuracy will improve over time."
Accuracy in disease mapping
“Genius AI Detection is particularly accurate and specific at delineating a lesion on the mammogram or tomosynthesis,” said Sharma. If there are multiple lesions, several mark-ups are shown to highlight specific lesions and the number of quadrants affected. She notes that accurate information on lesion number and location(s) prior to ultrasound enables multiple same-day biopsies and ultimately provides valuable information for the surgeon.
There is currently a drive to reduce the number of mastectomies performed in the U.K. and it has been shown that breast conserving surgery is now possible even for relatively extensive disease. However, disease mapping has to be accurate and the radiologist’s workup has significantly increased as a result.
“Other AI algorithms have not clearly delineated one versus multiple lesions, so this is a really positive aspect of Genius AI Detection,” Sharma commented. It has also been useful in mapping calcifications that are not typically visible on ultrasound, thus providing additional information for the biopsy.
Team impact
According to Sharma, the digital AI tool has been positively received by her team; the output is trusted by her colleagues who have received training to understand how the algorithm works and how to interpret the findings. During the initial service evaluation, all Genius AI Detection markups for multifocal lesions were found to be accurate compared with final histology and this provided the team with additional confidence in the technology.
The deep-learning AI tool is also proving valuable from a training perspective; results from the reader and algorithm can be compared to help improve reader performance. Sharma believes the implementation of AI is vital for the younger generation who should be able to train with and adopt AI early in their careers.
Encouraging radiologists who may be skeptical or cautious about integrating AI into their workflow, Sharma pointed out that Genius AI Detection has alerted her team to findings they may not have picked up on, for example, a larger area of breast may be affected than previously detected, and has provided reassurance of a normal result, thus helping to prevent an unnecessary biopsy.
“It is reassuring to work with a tool I can trust. When there is a prompt, it is usually due to an abnormality. This compares with the ImageChecker CAD where you would have 20 prompts for every one cancer detected,” she said.
The future of AI in breast screening
Sharma believes radiologists should engage with AI and communicate their specific needs so that effective solutions can be developed. “We are aiming to improve the patient experience and patient pathway, but this can only work if we also improve the clinical pathway by supporting clinicians to deliver the required workload. By embedding AI in our workflows, we can streamline our workload and ensure patients receive the most appropriate treatment in a timely way,” she explained.
Currently in the U.K., tomosynthesis and AI are not approved for the national breast screening program. In the U.S., Canada, and Europe, the type of breast screening offered depends on reimbursement; tomosynthesis is used first-line where screening is reimbursed, while 2D imaging is still offered where it is not.
In Sharma’s opinion, 3D tomosynthesis is the best imaging tool and should be the first-line screening option in the future when used with AI tools to accelerate workflow. Combined with 3D tomosynthesis, AI can reduce the number of images to be read. For example, 3DQuorum utilizes AI analytics to produce SmartSlices, then the use of Genius AI Detection accurately flags suspicious versus normal cases to improve cancer detection rate.
“With a dwindling workforce and increased workload, it is important that we create workflow efficiencies to avoid a backlog, while maintaining accuracy, and AI can help us with this,” she concluded.
Sharma and her colleagues will present findings from their study of Genius AI Detection v2.0 at the European Society of Breast Imaging annual scientific meeting later this year (25 to 27 September 2025, Aberdeen, U.K.). Consistent with the team’s clinical experience, the findings suggest that the deep-learning AI tool has high sensitivity and a high negative predictive value, with overall positive findings.