Standard imaging tests include MRI, CT, ultrasound, PET, and X-ray. Endoscopy—This procedure uses a specialized tool with a light or camera to look inside the body for a tumour. Biopsy—A sample of the patient’s tumour will be obtained and analyzed.
Some scientists are investigating the potential of artificial intelligence (AI). In a recent study, scientists trained an algorithm with encouraging results.
Artificial intelligence is currently gaining enormous importance in cancer medicine. But there are still problems, for example, when it comes to collaboration between humans and AI.
Artificial intelligence (AI) is extremely good at recognizing patterns. If you train it with thousands of cancer case studies, it will develop into an expert system for cancer detection. This system is equal to, if not superior to, human experts.
Skin cancer, breast and colon cancer, prostate and lung cancer: Computers now assist in diagnosing all common types of tumours. He relies on images from ultrasound, computer tomography, MRI, or the microscope, which is used to examine tissue samples.
Lack of transparency: Doctors do not yet trust AI systems
Yet the technology has a serious problem: most systems are not transparent. They do not explain how they arrived at their diagnosis. This means that doctors cannot compare the diagnoses with their specialist knowledge, which upsets Titus Brinker, who leads a working group on AI in cancer diagnostics at the German Cancer Research Center in Heidelberg.
“The doctor cannot understand how the system came to a decision. And that, in turn, leads to him not wanting to trust the system, not wanting to use it, and ultimately keeping AI out of the routine, even though it would make sense to integrate it .” Doctor Brinker’s team is working on diagnostic AI for skin cancer, which also explains how it arrived at its conclusion. Only then will humans and AI become a real team delivering the best diagnostic results possible. Brinker is convinced of this.
Too strict data protection stands in the way of AI use
But the dermatologist from Heidelberg points out another reason why AI-supported cancer diagnosis cannot develop its full potential in Europe: data protection. The European General Data Protection Regulation only allows the use of patient data under strict rules—for example, through anonymization. All characteristics and data that make a person identifiable are deleted, separated, or falsified. As a result, the AI is missing important general patient data that could make its diagnosis more accurate.
For physician Brinker, it is incomprehensible that data protection is more important than patient health. “Data protection is an issue for healthy people. Patient protection is currently under data protection. So data protection ultimately leads to us having much worse medicine in Europe.”
AI simplifies radiation therapy
Artificial intelligence is now widely used in tumordiagnostics. But there are also initial applications in tumor therapy. UrsulaNestle is chief physician in the radiation therapy department at the Maria HilfClinic in Mönchengladbach. In her field, she says, there is significantprogress through AI.
Until now, with radiation therapy often lasting several weeks, the radiation plan had to be readjusted for each individual treatment because the position of the organs in the patient’s body changes slightly from day to day.
Computer tomography is integrated into the latest radiation systems. It registers the current spatial conditions in the patient’s body in real-time and automatically adjusts the radiation plan with the help of AI. This means saving time, greater precision, and fewer side effects during radiation therapy.
AI-supported therapy plan: Patients have a say
Radiation therapist Nestle is also enthusiastically pursuing the development of an AI-supported patient information system. Tumour patients can go through various treatment options with their doctors based on scientific studies and personal patient data.
This allows patients and their therapists to make well-informed decisions about their radiation therapy. “There are systems where you can see, for example, if I do such and such treatment, I have such and such a chance, but also such and such a risk. And then perhaps there is an alternative or a different variant of this treatment.” This treatment has fewer side effects and, therefore, less tumour control, says Nestle.
Artificial intelligence is changing cancer medicine in many areas. However, experts like therapist Nestle also demand that clinical studies be conducted to examine how patients actually benefit from these innovations.
AI has identified 12% more cases of breast cancer in the UK for the first time.
A breast screening solution known as Mia, based on artificial intelligence (AI), has aided doctors in detecting 12% more cases of cancer than the typical procedure. This announcement was made today by Kheiron Medical Technologies, NHS Grampian, the University of Aberdeen, and Microsoft. If implemented across the entire NHS, a 12% increase in breast cancer detection could lead to improved outcomes for thousands of women in the UK. The enhanced AI workflow also demonstrated a reduction in the number of women unnecessarily called back for further assessment and projected a potential 30% decrease in workload.
Every year, over two million women undergo breast cancer screening in the UK, but detecting breast cancer is extremely challenging. Approximately 20% of women with breast cancer have tumors that go unnoticed by mammogram screening, which is why many countries require two radiologists to review every mammogram.
NHS Grampian, which delivers health and social care services to over 500,000 individuals in the North East of Scotland, carried out the initial formal prospective evaluation of Kheiron’s Mia AI solution (CE Mark class IIa) in the UK as part of a study involving 10,889 patients.
In this evaluation, funded by a UK Government ‘AI in Health and Care Award’, Mia helped medical personnel discover additional cases of cancer. The earlier identification of primarily high-grade cancers has allowed for earlier treatment, which is more likely to be successful. The evaluation also revealed no increase in the number of women unnecessarily recalled for further investigation due to false positives. As part of a simulated workflow with AI integration, a workload reduction of up to 30% was anticipated.
Barbara, from Aberdeen, was among the first women in the UK whose cancer was detected by Mia. Barbara stated, “My cancer was so small that the doctors said it would not have been detected by the naked eye.” Detecting her cancer at an earlier stage before it spread has provided Barbara with a significantly improved prognosis compared to her mother, who required more invasive treatment for her own breast cancer. She said, “It’s a lifesaver, it’s a life changer.”
Dr. Gerald Lip, who led the prospective trial at NHS Grampian, mentioned, “If cancer is detected when it is under 15mm, most women now have a 95% chance of survival. Not only did Mia help us identify more cases of cancer, most of which were invasive and high-grade, but we also projected that it could reduce the notification time for women from 14 days to just 3 days, reducing significant stress and anxiety for our patients.”
Professor Lesley Anderson, Chair in Health Data Science at the University of Aberdeen, remarked, “While our previous research, led by Dr. De Vries, suggested that Mia could identify more cases of cancer, the GEMINI trial results left us astounded. If Mia were utilized in breast screening, it would mean that more cases of cancer would be detected without subjecting more women to additional tests.”
“However,” she added, “our earlier research highlighted a potential issue – changes to the mammography equipment could impact Mia’s performance. To seamlessly integrate Mia into screening programs, we are collaborating closely with Kheiron to develop methods for monitoring and adjusting the AI, ensuring that it continues to deliver the impressive results we observed in the recent evaluation.”
“Receiving direct feedback from a woman whose cancer was picked up by Mia was a significant moment for everyone who has contributed to pioneering the development and evaluation of our AI technology,” said Peter Kecskemethy, CEO of Kheiron. “These outstanding results have surpassed our expectations, and we are immensely grateful to the teams from NHS Grampian, the University of Aberdeen, Microsoft, and the UK Government, who have enabled us to carry out this groundbreaking work.”
Identifiable patient data is removed before a mammogram is uploaded to the Azure Cloud. Once de-identified, the Mia software reads the mammogram and sends the recommendation back to the hospital or clinic. It is currently in use at 4 locations in Europe and 16 NHS sites in the UK as part of ongoing trials.
This large-scale deployment utilizing the Azure Cloud is part of the UK Government’s aim to be at the forefront of AI technology in healthcare. Representatives from Microsoft UK’s Healthcare and Life Sciences division believe that AI, in collaboration with medical professionals, can play a crucial role in improving patient outcomes, as evidenced by the results of the prospective evaluation at NHS Grampian. Thanks to this pioneering work, more women have an increased chance of overcoming cancer.
A team of researchers from Denmark and the Netherlands has combined an AI diagnostic tool with a mammographic texture model to enhance the assessment of short- and long-term breast cancer risk. This innovative approach represents a significant advancement in refining the ability to predict the complexities of breast cancer risk.
Approximately one out of every ten women will develop breast cancer at some point in their life. Breast cancer is the most prevalent type of cancer in women, with diagnoses predominantly occurring in women over the age of 50. Although current screening programs primarily use mammography for early breast cancer detection, some abnormalities can be challenging for radiologists to identify. Microcalcifications, which are tiny calcium deposits often no larger than 0.1 mm, are present in 55% of cases, and are either localized or broadly spread throughout the breast area.
These calcifications are commonly linked to premalignant and malignant lesions. Currently, the majority of breast cancer screening programs determine a woman’s estimated lifetime risk of developing breast cancer using standard protocols.
Dr. Andreas D. Lauritzen, PhD, from the Department of Computer Science at the University of Copenhagen in Denmark, noted that artificial intelligence (AI) can be employed to automatically detect breast cancer in mammograms and assess the risk of future breast cancer. Collaborating with researchers from the Department of Radiology and Nuclear Medicine at Radboud University, Nijmegen, in the Netherlands, Dr. Lauritzen and his team worked on a project that combined two types of AI tools to capitalize on the strengths of each approach: diagnostic models to estimate short-term breast cancer risk and AI models to identify breast density using mammographic texture.
A group of seven researchers from Denmark and the Netherlands conducted a retrospective study of Danish women to determine whether a commercial diagnostic AI tool and an AI texture model, trained separately and then combined, could enhance breast cancer risk assessment. They utilized a diagnostic AI system called Transpara, version 1.7.0, from the Nijmegen-based company Screenpoint Medical B.V., along with their self-developed texture model comprising the deep learning encoder SE-ResNet 18, release 1.0.
Dr. My C. von Euler-Chelpin, associate professor at the Centre for Epidemiology and Screening, Institute of Public Health, University of Copenhagen, stated that the deep learning models were trained using a Dutch training set of over 39,245 exams. The short- and long-term risk models were combined using a three-layer neural network. The combined AI model was tested on a study group of more than 119,650 women participating in a breast cancer screening program in the Capital Region of Denmark over a three-year period from November 2012 to December 2015, with at least five years of follow-up data. The average age of the women was 59 years.
Key findings from the study, which was published in Radiology and presented at the latest Radiological Society of North America (RSNA) annual meeting in Chicago in November 2023, revealed that the combined model achieved a higher area under the curve (AUC) compared to the diagnostic AI or texture risk models separately, for both interval cancers (diagnosed within two years of screening) and long-term cancers (diagnosed after this period).
The combined AI model also enabled the identification of women at high risk of breast cancer, with women in the top 10% combined risk category accounting for 44.1% of interval cancers and 33.7% of long-term cancers. Dr. Lauritzen and his colleagues concluded that mammography-based breast cancer risk assessment is enhanced when combining an AI system for lesion detection and a mammographic texture model. Using AI to assess a woman’s breast cancer risk from a single mammogram will lead to earlier cancer detection and help alleviate the burden on the healthcare system due to the global shortage of specialized breast radiologists.
Dr. Lauritzen expressed that the current advanced clinical risk models typically require multiple tests such as blood work, genetic testing, mammograms, and extensive questionnaires, all of which would significantly increase the workload in the screening clinic. Using their model, risk can be evaluated with the same precision as clinical risk models, but within seconds from screening and without introducing additional workload in the clinic, as mentioned in an RSNA press release.
The Danish-Dutch research team will now focus on investigating the combination model architecture and further ascertaining whether the model is adaptable to other mammographic devices and institutions. They also noted in their paper that additional research should concentrate on translating combined risk to lifetime or absolute risk for comparison with traditional models.
What is EBCD?
The Enhanced Breast Cancer Detection program utilizes artificial intelligence (AI) technology and a thorough clinical review process to improve areas of concern in screening mammography. Each step in the screening process is overseen by a certified radiologist. The final results of the patient’s examination are reported by the radiologist.
EBCD provides an extra layer of confidence in the examination results as it is similar to having multiple sets of eyes on the mammogram: the initial radiologist, the FDA-cleared AI, and an additional breast-specialty radiologist. This protocol has demonstrated the ability to discover 17% more cancers and can also aid in reducing recall rates.
AI for breast cancer detection: digital MMG and DBT
The increasing number of medical scans, shortage of radiologists, and the critical need for early and accurate cancer detection have emphasized the requirement for an improved CAD system, despite the limitations of traditional CAD systems. The rapid advancements in AI and DL techniques have created opportunities for the development of advanced CAD systems that can identify subtle signs and features that may not be immediately noticeable to the human eye.
The development of AI-CAD commences with the gathering of a large dataset representing the target population and imaging device. Human readers then collaborate to identify and label lesions in mammograms based on confirmed pathological reports for breast cancer detection. Utilizing these labeled images, AI-CAD self-learns the features used for training, which distinguishes it critically from traditional CAD, which only learns human-derived features. To enhance the algorithm’s performance, internal validation was conducted using a dataset separate from the training data to prevent overfitting.
The outcome is an AI-CAD system that can achieve high cancer detection rates while sustaining high specificity, and it performs significantly better than traditional CAD. This groundbreaking technology has the potential to enhance accuracy, boost efficiency, and reduce diagnostic variability in breast cancer screening. This can alleviate the workload on radiologists and facilitate timely and accurate diagnoses.
AI can be integrated into the workflow of 2D breast screening in various scenarios, including using AI as a standalone system to replace a human reader, and concurrent reading with AI-CAD or AI for triaging normal cases. In double-reading screening, AI may assume the role of a second reader or CAD for one or both readers.
Alternatively, AI can pre-screen normal cases and reduce the workload for radiologists, or employ a rule-in rule-out approach to remove low-risk cases and refer high-risk cases for another reading by radiologists. When deciding how AI will be integrated into a workflow, factors such as target sensitivity, specificity, recall rate, and reading workflow in the target country must be taken into account. Stand-alone AI performance was evaluated to simulate a scenario in which AI entirely replaces a human reader. Several studies have shown that AI can perform as well as or even better than humans. According to a systematic review and meta-analysis of 16 studies, standalone AI performed equally well or better than individual radiologists in digital MMG interpretation, based on sensitivity, specificity, and AUC metrics.
AI also surpasses radiologists in DBT interpretation, but further evidence is needed for a more comprehensive assessment. This emphasizes the potential of AI in independent mammographic screening, which is particularly significant for countries that employ double reading, as replacing a human reader with AI can result in significant reductions in required human resources.
Selecting an optimal AI output score, known as the threshold score or operating point, is crucial for the implementation of AI algorithms for diagnostic decision-making. While AI algorithms often have a default threshold score, it is essential to recognize that different scenarios may require different scores. Factors such as the specific workflow in which the AI was used or the goals of the screening program should be considered when determining the most suitable algorithm threshold score.
For instance, Dembrower et al. compared the sensitivity and workload of standalone AI versus a combination of AI and radiologist. When the sensitivity of the standalone AI was matched with that of a human radiologist, it demonstrated a potential relative sensitivity approximately 5% higher than that for the combined sensitivity of the AI and radiologist, also matching that of the two radiologists.
However, the workload involved in the consensus discussions for the standalone AI scenario was nearly double that of the combined AI reader approach. This suggests that the combined AI-reader scenario and associated AI algorithm threshold may be more suitable for screening programs aimed at reducing the workload while maintaining similar sensitivity compared to having two readers.
In a different reader study for DBT, it has also been noted that the use of AI not only improved the performance of radiologists (0.795 without AI to 0.852 with AI) but also decreased the reading time by up to 50% (from 64.1 seconds without AI to 30.4 seconds with AI).
AI triage is another technique for evaluating AI algorithms. Since most screening mammograms show no signs of malignancy, even removing a portion of normal exams can significantly reduce the workload. Dembrower and colleagues demonstrated that AI can be set at a threshold where 60% of cases can be safely removed from the worklist without risking missing cancer cases.
Similar results have been reported in other studies, with a 47% reduction in workload resulting in only 7% missed cancers. Furthermore, a “rule-in” approach can be utilized, where cases labeled as benign by human readers but assigned a high score by AI are automatically recalled for further testing. This combined approach can effectively reduce the workload while increasing the detection of subsequent interval cancers (ICs) and next-round detected cancers.
Retrospective studies utilize existing data representing target populations and allow various simulations to test AI algorithms. Radiologists’ decisions and histopathological data were required for comparison. It is common practice to establish the ground truth based on at least two consecutive screening episodes to detect screen-detected cancers, ICs, and next-round detected cancers. Promising results have been achieved; however, most retrospective studies are limited to validating AI algorithm performance in an enriched cohort or multiple-reader multiple-case analysis.
An area of recent interest in AI cancer-detection algorithms is improving the detection of ICs. ICs are often aggressive forms of cancer associated with higher mortality rates, and the risk of death from IC is 3.5 times higher than that of non-ICs. Despite previous efforts, IC accounts for approximately 30% of detected breast cancers, and attempts to improve IC detection have been unsuccessful. However, AI algorithms have shown promise in detecting ICs. Hickman and colleagues demonstrated that a standalone AI can detect 23.7% of ICs, even when set at a 96% threshold, potentially allowing for a significant increase in IC detection.
With substantial retrospective evidence available, ongoing efforts are being made worldwide to conduct prospective clinical trials. Results from several prospective trials investigating the use of AI in 2D breast screening are emerging. For example, the ScreenTrustCAD study conducted in Sweden examined the impact of replacing one reader in a double-reading setting. The results were highly positive, indicating that in a prospective interventional study based on a large population, a single reader with AI can achieve a superior cancer detection rate while maintaining the recall rate compared with traditional double readers.
In this scenario, the effects of AI on arbitration can only be prospectively evaluated. In another RCT conducted in Sweden, called the Mammography Screening with Artificial Intelligence trial, the clinical safety of using AI as a detection support in MMG screening was investigated. In an intervention group, examinations were first classified by AI into high- and low-risk groups, which were then double- or single-read, respectively, by radiologists with AI support.
Interim analysis results showed that AI-supported screening not only demonstrated comparable cancer detection rates to a control group’s standard double reading but also significantly reduced screen-reading workload. This RCT indicated that employing AI in MMG screening could be a safe and effective alternative to standard double reading in Europe. The trial will continue for two more years to assess the primary endpoint of the IC rate. Other studies, such as the AI-STREAM in South Korea, are also actively investigating the effects of AI in single-reader concurrent reading settings.
Prospective trials are indeed crucial, as they provide valuable insights into the performance of AI algorithms in real clinical settings and capture the challenges that may arise in these environments. A pitfall of retrospective trials is that they often use cancer-enriched datasets that do not reflect the real-life prevalence of cancer. Therefore, AI performance from these skewed studies may not necessarily be replicated in prospective studies or real life.
Prospective trials, on the other hand, allow the evaluation of AI algorithms in out-of-distribution scenarios, providing a more realistic assessment of their performance. However, the disadvantage of prospective studies is their high cost and lengthy time frame, which makes it difficult to conduct them frequently.
In a different reader study for DBT, it was also noted that the use of AI not only improved the performance of radiologists (from 0.795 without AI to 0.852 with AI) but also reduced the reading time by up to 50% (from 64.1 seconds without AI to 30.4 seconds with AI).
AI triage is another technique for testing AI algorithms. Since most screening mammograms show no signs of malignancy, eliminating even a portion of normal exams can significantly reduce the workload. Dembrower et al. demonstrated that AI can be set at a threshold where 60% of cases can be safely removed from the worklist without risking missing cancer cases.
Similar results have been reported in other studies, with a 47% reduction in workload resulting in only 7% missed cancers. Additionally, a “rule-in” approach can be used where cases labeled as benign by human readers but assigned a high score by AI are automatically recalled for further testing. This workflow, combined with the “rule-out” approach, can significantly reduce the workload while increasing the detection of subsequent interval cancers (ICs) and next-round detected cancers.
Retrospective studies use existing data representing target populations and allow various simulations to test AI algorithms. The decisions of radiologists and histopathological data were necessary for comparison. It is common practice to establish the ground truth based on at least two consecutive screening episodes to detect screen-detected cancers, ICs, and next-round detected cancers. Promising results have been achieved; however, most retrospective studies are limited to the validation of AI algorithm performance in an enriched cohort or multiple-reader multiple-case analysis.
A recent area of interest in AI cancer-detection algorithms is the improvement of IC detection. ICs are often aggressive forms of cancer associated with higher mortality rates, and the risk of death from IC is 3.5 times higher than that of non-ICs. Despite previous efforts, IC accounts for approximately 30% of detected breast cancers, and attempts to improve IC detection have been unsuccessful. However, AI algorithms have shown promise in detecting ICs. Hickman et al. demonstrated that a standalone AI can detect 23.7% of ICs, even when set at a 96% threshold, potentially allowing for a significant increase in IC detection.
With the abundance of available retrospective evidence, ongoing efforts are being made worldwide to conduct prospective clinical trials. Results of several prospective trials investigating the use of AI in 2D breast screening are emerging. For example, the ScreenTrustCAD study conducted in Sweden examined the impact of replacing one reader in a double-reading setting. The results were highly positive, showing that in a prospective interventional study based on a large population, a single reader with AI can achieve a superior cancer detection rate, while maintaining the recall rate compared with traditional double readers.
In this situation, the effects of AI on arbitration can only be prospectively evaluated. In another RCT conducted in Sweden called the Mammography Screening with Artificial Intelligence trial, the clinical safety of using AI as a detection support in MMG screening was investigated. In an intervention group, examinations were first classified by AI into high- and low-risk groups, which were then double- or single-read, respectively, by radiologists with AI support.
Interim analysis results showed that AI-supported screening not only showed comparable cancer detection rates to a control group’s standard double reading but also significantly reduced screen-reading workload. This RCT revealed that employing AI in MMG screening could be a safe and effective alternative to standard double reading in Europe. The trial will continue for two more years to assess the primary endpoint of the IC rate [64]. Other studies, such as the AI-STREAM in South Korea, are also actively investigating the effects of AI in single-reader concurrent reading settings.
Prospective trials are indeed essential, as they provide valuable insights into the performance of AI algorithms in real clinical settings and capture the challenges that may arise in these environments. A pitfall of retrospective trials is that cancer-enriched datasets that do not reflect the real-life prevalence of cancer are often used. Therefore, AI performance from these skewed studies may not necessarily be replicated in prospective studies or real life.
Prospective trials, on the other hand, enable the evaluation of AI algorithms in out-of-distribution scenarios, providing a more realistic assessment of their performance. However, the disadvantage of prospective studies is their high cost and lengthy time frame, which makes it challenging to conduct them frequently.
A possible solution for addressing the difficulties of conducting prospective trials for every use case and geographical area is to utilize large-scale retrospective studies using extensive datasets. These retrospective studies can take into account the variability encountered in real-life scenarios by collecting a sufficient sample size and integrating data from multiple centers.
National initiatives, such as the Swedish Validation of Artificial Intelligence for Breast Imaging project, demonstrate this approach by establishing comprehensive multicenter databases for external validation. This allows independent and simulated testing of AI algorithms. Combining insights from prospective and retrospective trials can ensure the cost-effectiveness, scalability, and safe adoption of AI in breast screening, benefiting both patients and healthcare systems.
AI is employed in supplemental breast cancer screening utilizing MRI/ultrasound. Additional imaging techniques, including DBT, MRI, handheld ultrasound, and automated breast ultrasound (ABUS), are commonly used in addition to traditional MMG for improved cancer detection in women with dense breasts. Efforts have been made to apply AI to these modalities to enhance their performance.
For example, Shen et al. showed that the implementation of an AI system improved the diagnostic process for identifying breast cancer using ultrasound. The use of AI resulted in a significant reduction in false-positive rates by 37.3% and biopsy requests by 27.8%, while maintaining sensitivity. Furthermore, a standalone AI system outperformed an average of ten board-certified BRs, with an AUROC improvement of 0.038 (95% CI, 0.028–0.052; p < 0.001). This implies that the AI system not only assists radiologists in improving the accuracy, consistency, and efficiency of breast ultrasound diagnosis but also performs better than human experts.
AI algorithms focusing on MRI enhancement aim to improve acquisition time, a critical issue in this modality. The ‘Fast MRI challenge’ is a research initiative aimed at developing and evaluating MRI techniques using AI to expedite MRI image acquisition without compromising image quality. Results from this challenge have demonstrated that AI can effectively reconstruct missing data in accelerated magnetic resonance images while maintaining acceptable data quality for radiologists.
Finally, as CAD systems, AI algorithms have proven to be useful in conjunction with supplemental imaging techniques. CAD-ABUS helps radiologists achieve a significant reduction in reading time while maintaining accuracy in detecting suspicious lesions. Additionally, in the case of MRI, DL-based CAD systems have shown a significantly higher average sensitivity in early phase scans where abbreviated MRI protocols are used. This underscores the potential of AI in playing an increasingly important role in the future, particularly in the interpretation of supplemental images.
Artificial intelligence can detect breast cancer in mammograms as effectively as experienced radiologists, according to a new study that some experts are calling a game changer for the field of oncology. The emerging technology could reduce radiologists’ workload by about half, allowing them to focus on more advanced diagnostic work, the study found.
The preliminary analysis of a long-term trial of 80,000 women in Sweden, published Tuesday in the journal Lancet Oncology, indicated that AI readings of mammograms actually detected 20 percent more cases of breast cancer than the “standard” reading by two radiologists. The AI assessments were verified by one or two radiologists, depending on the patient’s risk profile.
This led the researchers to conclude that using AI in mammography screening is a “safe” way to help reduce patient waiting times and ease the pressure on radiologists amid a global workforce shortage.
It may be some time before mammograms will be interpreted by a machine, as the authors and other experts have cautioned that AI models need further training and testing before being deployed in healthcare settings.
Nevertheless, the findings are “astonishing,” wrote Nereo Segnan and Antonio Ponti, experts associated with the Center for Epidemiology and Cancer Prevention in Turin, Italy, who were not involved in the analysis.
In an article accompanying the study release, they propose that integrating AI in screening procedures could ultimately lead to “reduced breast cancer mortality” by ensuring earlier identification of breast cancer, when it is more treatable. Given that breast cancer is the “world’s most prevalent cancer,” according to the World Health Organization, this would be a significant achievement.
The analysis is “groundbreaking,” according to Robert O’Connor, director of Ireland’s National Clinical Trials Office (NCTO), who wrote on X, formerly known as Twitter. It demonstrates that AI could aid in categorizing mammograms based on cancer risk and identify breast cancer in those mammograms at a higher rate than radiologists with at least a couple of years of experience.
Using machine learning to enhance medical diagnostics has been a longstanding practice, but it has gained momentum in recent years due to advancements in artificial intelligence.
The results of this research align with emerging studies indicating that AI has the potential to assist humans in detecting cancer earlier and more accurately, potentially leading to improved outcomes for patients. According to the authors, this is the first randomized controlled trial to explore the use of AI in mammography screening.
The trial enlisted 80,020 women aged 40 to 80 who underwent mammograms in Sweden between April 2021 and July 2022. Half of them were randomly selected to have their mammograms interpreted by a commercially available AI model alongside one or two radiologists, based on the risk score assigned by the AI during an initial screening. The other half had their mammograms assessed by two radiologists, which is considered the standard practice in Europe.
In addition to interpreting mammograms, the AI model provided radiologists with information from the initial screening to aid in accurate interpretation. Women with suspicious mammograms were asked to undergo further tests.
Overall, the AI-supported screenings detected breast cancer in 244 women, compared to 203 in the standard screening group, representing a 20 percent difference.
Improving the detection rates of breast cancers is crucial, as early-stage breast cancers are increasingly treatable.
In 2020, the disease claimed the lives of at least 685,000 women worldwide, according to the WHO. The average woman in the United States has a 13 percent chance of developing breast cancer in her lifetime, with a roughly 2.5 percent chance of dying from the disease, as stated by the American Cancer Society.
The study found that AI-supported screenings did not result in higher rates of false positives.
While the authors did not measure the time taken by radiologists to interpret the mammograms, they estimated that a single radiologist would have taken 4 to 6 months less to read the mammograms in the AI test group compared to those in the standard screening group, assuming a rate of about 50 readings per hour per radiologist.
James O’Connor, a professor of radiology at the Institute of Cancer Research in London, believes that integrating AI into breast cancer screenings could significantly impact the daily work of professionals in the field.
If AI-supported screenings can be implemented across different jurisdictions and populations, and be accepted by patients, regulators, and healthcare professionals, there is potential to save a significant amount of time and help alleviate workflow shortages, according to O’Connor. However, he acknowledges that questions remain around the implementation of AI in medical care, particularly due to varying regulations across different countries and potential patient concerns.
James O’Connor dismissed the idea of artificial intelligence replacing radiologists as “nonsense.” Instead, he highlighted the potential for the right AI model, if properly implemented, to assist radiologists in focusing on challenging cases and other types of scans.
The lead author of the study, Kristina Lang, expressed in a news release that while the interim safety results are promising, they are not sufficient on their own to confirm the readiness of AI to be implemented in mammography screening.
A concern arising from the study is that while AI-supported screenings detected more cancers, they may also lead to overdiagnosis or detection of cancers that pose a low risk to patients.
During the study, screenings aided by AI identified more “in situ” cancers, which are cancerous cells that have not yet spread and may turn out to be low-grade. The authors noted that this could potentially lead to over-treatment of conditions that may not necessarily pose a threat, including through procedures such as mastectomies.
Furthermore, the study did not gather data on the race and ethnicity of the patients, so it cannot determine whether AI-supported screenings are more effective in identifying cancers in particular demographic groups.
Robert O’Connor of the NCTO pointed out the importance of validation in multiple countries due to variations in the presentation of breast cancer among different ethnicities and age groups.
According to research, artificial intelligence has the potential to reduce significantly the number of missed early-stage breast cancer cases and enhance medical diagnoses, demonstrating the technology’s capability to improve and expedite the process.
AI analysis identified up to 13 percent more cases than those diagnosed by doctors, which is a substantial proportion of the 20 percent or more cancers estimated to be overlooked using current non-AI screening methods.
A new research paper, which was published in Nature Medicine on Thursday, demonstrates the potential of machine learning in addressing life-threatening threats by identifying errors or detecting subtle signs that may be overlooked by human observers.
Ben Glocker, a professor specializing in machine learning for imaging at Imperial College London and one of the study’s co-authors, emphasized the significance of using AI as a safety net to prevent subtle indications of cancer from being overlooked. He stated, “Our study shows that using AI can act as an effective safety net — a tool to prevent subtler signs of cancer falling through the cracks.”
The researchers used an AI tool called Mia, which was developed by Kheiron Medical Technologies, a UK-based company specializing in AI medical diagnostics. The study focused on 25,000 women who underwent breast cancer screening in Hungary between 2021 and 2023.
The study consisted of three phases, each involving different interactions between radiologists and the AI. The groups showed improvements in cancer detection rates of 5 percent, 10 percent, and 13 percent, compared to the standard reading by at least two radiologists.
The additional cancers detected were mainly invasive, indicating their potential to spread to other parts of the body.
These findings provide important evidence that AI can enhance the accuracy and speed of identifying malignant tissues. A study from Sweden, published in late August, also showed similar cancer detection rates between AI-enhanced analysis of mammograms and standard human double reading.
Dr. Katharine Halliday, president of the UK’s Royal College of Radiologists, acknowledged the potential of AI to speed up diagnosis and treatment, calling the research from Hungary “a promising example of how we can utilize AI to speed up diagnosis and treatment” in the NHS.
The use of AI also offers the possibility of expediting analysis. The authors of the Hungarian paper mentioned that Mia could potentially save up to 45 percent of the time spent on breast cancer scan reading times.
Kheiron Medical Technologies reported that Mia has been piloted at 16 hospitals in the UK and is being introduced in the US.
The researchers stressed the importance of further expanding and deepening the application of AI in cancer detection. They highlighted the need to gather results from more countries, utilizing other AI systems, and monitor the emergence of additional cancer cases in their study group.
In GlobalData’s Clinical Trials Database, there are presently 1,490 ongoing clinical trials for in vitro diagnostics (IVD) devices, with 569 of those trials dedicated to oncology diagnostic devices. Specifically, nine of these trials focus on analysis or partial analysis.
This month, Mindpeak, a provider of artificial intelligence (AI) solutions and software, formed a partnership with Proscia, a company specializing in computational and digital pathology solutions, to enhance cancer diagnosis. The collaboration aims to optimize pathologists’ workflows using AI, allowing for more efficient clinical decisions based on digital pathology images from patient samples. The objective of this partnership is to utilize Mindpeak’s breast cancer detection software, BreastIHC, alongside Proscia’s open digital pathology platform, Concentriq Dx, to improve breast cancer diagnosis through AI-driven digital pathology analysis.
Additionally, an active trial called Artificial Intelligence Neuropathologist, conducted by Huashan Hospital and United Imaging Healthcare, is evaluating the capacity of their AI to identify central nervous system (CNS) tumors in an unsupervised and fully automated manner. This development is intended to enable quicker treatment for patients, as the device analyzes and processes samples more rapidly than physicians, enhancing diagnostic accuracy.
The aim of this trial is to create a self-learning AI device capable of achieving a clinical pathological diagnosis accuracy of 90% or higher.
With these innovative devices on the horizon, GlobalData anticipates that in the upcoming decade, a greater number of IVD manufacturers will incorporate AI technology into their devices to enhance diagnostic and treatment predictions, as well as oncologists’ workflows. Consequently, more individuals will have the opportunity to receive life-saving interventions at earlier stages of cancer, along with treatments that AI has shown to be the most effective.
Leave a Reply