10 insightful questions about AI answered by our experts
In this arcticle, we bring together two unique perspectives united by the shared mission of improving the future of breast cancer screening. We gathered questions from several women, including some breast cancer survivors, and asked Dr Mehran Taghipour-Gorjikolaie, our own leading scientist from LSBU who specializes in AI, to answer them. The result is an insightful conversation that delves into the hopes, challenges, and real-world impact of AI in oncology and explains how it is transforming cancer diagnosis, treatment, and patient support. We hope to contribute to the dialogue between people with lived experience and scientists that is shaping innovation in healthcare under the Horizon Europe programme. MammoScreen (MS): How does AI help doctors make better decisions about diagnosis and treatment? Dr. Mehran Taghipour-Gorjikolaie (MTG): AI assists doctors by rapidly analysing vast amounts of complex medical data to identify subtle patterns that might be difficult for humans to detect. For instance, in the context of breast cancer detection, AI can analyse microwave signal responses from breast tissue and highlight anomalies that may indicate the presence of a tumour. By providing data-driven insights and enhancing pattern recognition, AI supports clinicians in making more accurate, timely, and personalized diagnostic and treatment decisions. MS: What type of medical data does AI analyse, and how does it learn from it? MTG: AI can analyse a wide range of medical data, including medical images (such as MRIs, CT scans, and mammograms), physiological waveforms (like ECGs or microwave signal responses), laboratory test results, and even unstructured data such as clinical notes and patient records. It learns through a process known as supervised learning, where it is trained on large datasets containing labelled examples, by analysing thousands – or even millions – of cases with known outcomes (such as confirmed labels). MS: Can AI detect diseases earlier or more accurately than traditional methods? MTG: It has been demonstrated that it can do, in many cases. For instance, in the case of breast cancer, traditional imaging might miss tumours in dense breast tissue. AI that analyses microwave signals, as we are developing in MammoScreen, can offer a different type of insight, potentially detecting small or early-stage tumours that aren’t yet visible on a mammogram. Early results in research are certainly promising, but for the moment AI is only used to support, not replace, traditional diagnostic methods. MS: How accurate are AI tools compared to human doctors? MTG: AI tools can match or sometimes exceed human performance in specific tasks, like spotting signs of disease in images. However, they are most powerful when combined with expert review. For example, AI might catch a tiny signal that a human might miss, but the doctor adds context (such as patient history) to decide what it really means. MS: How can we ensure that AI doesn’t make mistakes or miss important details? MTG: Ensuring the safety and reliability of AI in healthcare requires a multi-layered approach. First, AI systems must be trained on large, diverse, and high-quality datasets that represent the full spectrum of patient populations and conditions. Rigorous validation and testing are essential before deployment in clinical settings. Additionally, AI models can be designed to flag cases where their confidence is low, prompting human review rather than automated decisions. Continuous performance monitoring, regular updates, and real-world audits are also critical to detect and correct any errors or biases early. Ultimately, AI should be seen as a tool that supports – not replaces – clinical expertise. MS: Can AI explain its decisions or is it a “black box” that doctors must trust blindly? MTG: This is one of the key challenges in integrating AI into clinical practice. While some AI models – particularly deep learning systems – have traditionally functioned as “black boxes,” offering little insight into how decisions are made, the field is rapidly advancing toward greater transparency. Increasingly, researchers are developing explainable AI (XAI) techniques that highlight which features of a signal, image, or dataset contributed most to a particular decision. These tools can, for example, show specific areas of a medical image that influenced the AI’s diagnosis. This interpretability is crucial for building clinicians’ trust, supporting informed decision-making, and ensuring AI serves as a transparent, accountable partner in patient care. MS: How can AI improve my personal healthcare experience and treatment outcomes? MTG: AI can lead to faster diagnoses, fewer unnecessary tests, and more personalized treatment plans. In breast cancer screening, for instance, AI may reduce false alarms (which cause stress) and catch cancers earlier, which can improve survival rates and quality of life. MS: What happens if an AI system gives a recommendation that conflicts with a doctor’s opinion? MTG: Doctors always have the final say. AI provides another perspective, but it’s the doctor’s job to decide what’s best for the patient. AI-based models work as an assistant for doctors, and they can’t make a final decision. Moreover, some systems can trigger a second review or deeper investigation in cases of disagreement. MS: Are there risks to relying on AI in medicine, and how are they managed? MTG: While AI has the potential to greatly support clinical decision-making, it is still too early to rely on AI models without human oversight. There are inherent risks – such as misdiagnosis, overfitting to specific datasets, or failing to generalize across diverse populations. These risks can be mitigated through several key strategies: training AI models on large, diverse, and high-quality datasets; integrating information at multiple levels – from raw sensor data to final decision-making stages (a process known as multi-level data fusion); and rigorously testing the models on previously unseen “blind” datasets to assess real-world performance. Fine-tuning and continuous monitoring are also essential to ensure safe and reliable operation. Importantly, AI should be viewed as a clinical decision support tool – meant to assist, not replace, healthcare professionals. MS: How do you see AI evolving in healthcare, and will it ever replace doctors? MTG: AI will keep getting better at helping with diagnosis, planning treatment, and even predicting future health
Be aware of hereditary breast cancer
Certain DNA mutations can significantly increase the risk of breast cancer and individuals with a first-degree relative (parent, sibling, or child) diagnosed with breast cancer may be at a higher risk of developing the disease when compared to the general population. While there’s no guaranteed way to prevent breast cancer, certain factors can be managed to reduce risk, and early detection significantly improves treatment outcomes. If you are worried about a possible family history of breast cancer, we recommend the following: 1. Learn more about your family health history: Find out if close relatives (parents, siblings, grandparents, aunts/uncles) have had breast or ovarian cancer, especially at a young age. 2. Talk to a Doctor: A healthcare professional can assess your risk based on your family history and recommend the next steps. 3. Consider genetic testing: If multiple family members have had breast cancer, a doctor may suggest genetic testing for mutations in genes like BRCA1 or BRCA2. 4. Get screened regularly: If you’re at high risk, you may need earlier or more frequent screenings (such as mammograms, MRI scans, or microwave imaging). 5. Discuss preventive options with your doctor: Some medication can lower breast cancer risk and preventive surgery (like a mastectomy) may be an option. 6. Maintain a healthy lifestyle: Exercise, eat well, avoid smoking and limit alcohol. A healthy lifestyle can reduce overall cancer risk even if you have a genetic predisposition. For more information about hereditary cancer, visit the EVITA Association website where you will find information about common hereditary cancer syndromes, a glossary of technical terms and additional downloadable support materials. EVITA has recently launched a digital platform aimed at fostering collaboration between citizens, the medical community and scientific researchers, with a focus on hereditary cancer. To contribute to research or to connect with others, visit EVITA Platform and register to join the community.
The MammoScreen interim analysis
One of the activities planned within the scope of the MammoScreen clinical investigation is the interim analysis, a data checkpoint and an opportunity to evaluate how well the AI algorithm is performing before the end of the study. The interim analysis was planned for the moment when the clinical trial reached 3,000 women, and we are pleased to have achieved this pivotal milestone in the beginning of 2025, as estimated in the recruitment projections. With the interim analysis, our researchers want to check if the MammoWave technology combined with AI is correctly identifying signs of breast cancer. For that, they look at key performance indicators, such as sensitivity and specificity, to ascertain if the AI routine is detecting cancer when it’s there and if false alarms are being reduced. One of the key drivers for the interim analysis is to ensure full compliance with our rigorous patient safety protocols and ethical standards. The team also needs to be able to identify areas for improvement before the end of the study, so that corrective action can be implemented. For instance, if the analysis shows that certain patterns are being misinterpreted, our experts can refine the AI algorithm to improve accuracy. On another hand, if the preliminary data shows that the technology is underperforming or leading to too many false results, adjustments could be made to improve the monitored parameters. Overall, the interim analysis is a powerful guide for future development and the results will help researchers decide what needs fine-tuning in the AI data analysis. The interim analysis is a critical milestone. It helps to validate and refine the AI model and ensures that the project stays on track. The clinical team will resume the recruitment of participants (aiming for up to 10,000 women) once the interim analysis is completed, using what they learned to update and improve the AI model before the final analysis. This step-by-step approach ensures that the technology is as accurate and safe as possible before being considered for wider use.
LSBU’s research contributions to the MammoScreen Project
Led by Dr. Mohammad Ghavami, Professor of Telecommunications Engineering at London South Bank University (LSBU), the LSBU research team plays a key role in the development and validation of MammoWave®, the cutting-edge microwave imaging device being tested in our clinical trial. Beyond MammoWave, the team´s research has contributed to the broader field of microwave imaging, particularly through work on ultra-wideband (UWB) technologies. Research into the dielectric differences in biological tissues provides a solid foundation for designing safer and more efficient diagnostic tools for medical applications, including cancer and bone imaging. Working within the School of Engineering, the team focuses on optimizing microwave imaging technologies in general and is working in many other innovative applications. The impact of this work has been amplified through major research grants and industry collaborations, including two EU-funded Marie Skłodowska-Curie Actions, WEBOING and ROVER, which explore the use of UWB radar in biomedical contexts. But their partnerships also extend to the development of technologies like smart card antennas. One of the team’s major contributions has been the clinical and phantom validation of MammoWave. In a study involving 51 breast scans, they demonstrated the device’s ability to detect lesions by mapping the tissue´s dielectric properties and analysing their homogeneity. These early results paved the way for MammoWave to be investigated as a non-invasive, radiation-free breast imaging method with potential applicability for cancer screening in populations who may benefit from this alternative technology. To enhance MammoWave’s diagnostic performance, the LSBU team is also responsible for integrating AI into the image analysis process, using unsupervised clustering, which is a type of AI tool that helps a computer find patterns in data without being told what to look for. The researchers successfully categorized over 1,000 samples from two European hospitals, achieving around 70% accuracy in distinguishing healthy from non-healthy cases. This AI-enhanced approach helps overcome challenges associated with data variability and increases the system’s ability to generalize across different patient populations. Looking ahead, the LSBU team’s ongoing research and integration of AI and machine learning are expected to further improve the accuracy and clinical relevance of MammoWave. Their pioneering efforts are helping to deliver innovative, patient-friendly alternatives to traditional diagnostic methods and their work fully exemplifies the potential of combining advanced engineering principles with medical diagnostics to improve cancer detection.
How is a MammoWave® report viewed and interpreted
MammoWave®, the technology being evaluated in the MammoScreen Project, uses microwave signals rather than X-rays to detect breast abnormalities. It operates in the 1 – 9 GHz frequency range and is non-invasive, does not emit ionizing radiation and requires no breast compression, making it safer and more comfortable for patients compared to traditional mammography. The device utilizes ultra-wideband (UWB) signals to differentiate between healthy and malignant tissues based on their dielectric properties, which can be understood as the way a material stores and transmits electric energy. It works in the detection of cancer, because every material – including human tissue – reacts differently when exposed to electromagnetic waves. Preliminary studies have shown that MammoWave® can detect subtle differences between normal tissue and cancer and MammoScreen is testing the device on a larger scale to gather further supporting evidence. The output of a MammoWave® scan is not a conventional image. Instead, it is a graphical representation of how tissue affects microwave signals. These patterns form the basis of a unique type of report that includes 2D images showing variations in tissue properties as shown in the diagram below, as well as a summary of dielectric measurements across different breast regions and an AI-based risk assessment, indicating whether the scan appears normal or if suspicious findings need further investigation. According to the protocol of our clinical study, if the AI label indicates a suspicion of malignancy, an additional ultrasound is performed to further investigate a potential abnormality, with particular attention devoted in the case of dense breast tissue. The healthcare professional will consider additional factors, such as previous imaging results, clinical symptoms, and patient history, to arrive at a final assessment. MammoWave® provides a new tool for early breast cancer detection that prioritizes patient comfort and safety and MammoScreen is collecting data to support its clinical validation. The strategy of combining AI and non-ionizing microwave imaging represents a departure from traditional mammography assessment methods that, we hope, will be an important step towards more accessible, inclusive and accurate breast cancer screening.
Developing AI for early breast cancer detection
The MammoScreen Project is improving the AI-driven algorithm used in the MammoWave® technology. The aim is to develop the AI-based model so that it detects breast cancer in an early stage while avoiding false results, but this is a fine balance to attain. The strategy for developing AI methods and ensure robust clinical validation involves an iterative process of data acquisition, improvement of algorithms and rigorous testing of the results against current clinical standards. The first step involves collecting high-quality and representative datasets. That is why MammoScreen gathers microwave imaging data from multiple clinical sites, recruiting across different patient demographics to ensure that variations in breast tissue composition are sufficiently represented in the dataset. Yet, before being used at all, the data needs to be checked for quality and pre-processed in order to reduce signal noise and remove artifacts (for example, irrelevant signals) that could interfere with the AI learning. The analysis also requires annotations regarding the presence or absence of tissue anomalies according to a radiologist´s assessment, so that supervised learning, a type of algorithm that learns from labelled data, can be applied. The output of the AI model is then compared to the true label and parameters are adjusted to reduce errors. The ultimate goal is to optimise the algorithm so that it can be used to predict the correct answer for new samples. A key component of the current AI-driven framework is a hierarchical model that employs a structured, multi-stage decision-making process. This stepwise approach systematically refines predictions, leading to significant improvements in the accuracy of breast cancer detection. In our new approach, we have begun exploring the use of deep learning strategies to enhance diagnostic accuracy. Deep learning, a subset of machine learning, employs multi-layered artificial neural networks inspired by the architecture of the human brain. These networks are capable of automatically learning and extracting hierarchical representations from complex data. Each layer within the network captures progressively abstract and informative features, enabling the model to identify intricate patterns that may not be easily discernible through traditional methods. In the context of microwave imaging, deep learning models can be trained to recognize spatial and spectral patterns—such as contours, shapes, textures, and intensity variations—correlated with the presence or absence of suspicious lesions. This capability holds significant promise for improving early breast cancer detection through automated and data-driven analysis. Our experts have been refining the MammoWave AI algorithm ahead of the planned interim analysis and the results are very promising. The refinement strategy involves the selection and enhancement of microwave signal features that best differentiate normal and abnormal tissues. A rigorous evaluation is being conducted to ensure that the AI model minimizes false results. The focus has been in transparency in what is done by the AI model, keeping in mind that radiologists will need to understand the rationale behind the AI-driven recommendations. By refining AI methods and validating clinical efficacy, the MammoScreen project aims to offer a reliable, non-invasive alternative for early breast cancer detection. Clinical validation is crucial when translating MammoScreen’s AI methodology into real-world healthcare settings.
