AI is revolutionizing many industries, including healthcare, where it now plays important roles in research, drug discovery, and clinical care. One area where AI is becoming particularly transformative is cancer diagnostics. AI is used to detect cancers from imaging tests like magnetic resonance imaging (MRI) and computed tomography (CT). In this context, AI facilitates faster and more accurate image analysis, leading to more efficient diagnoses and better patient care.
Google’s recent advancements in AI technology have shown impressive results in detecting breast cancer. This is an exciting development, as early detection of breast cancer is critical for improving patient outcomes and survival rates. However, whether Google’s AI can outperform human radiologists remains a complicated question to answer.
This article will cover the recent advances in AI, how it is used to detect breast cancer, and the advantages and challenges of using AI for cancer diagnostics.
Google’s recent advancements in AI technology have shown impressive results in detecting breast cancer. This is an exciting development, as early detection of breast cancer is critical for improving patient outcomes and survival rates. However, whether Google’s AI can outperform human radiologists remains a complicated question to answer.
Scientists train AI to distinguish between healthy and abnormal breast tissue using mammograms (images generated from mammography exams). This involves feeding AI systems on thousands of mammogram images that trained radiologists have already analyzed and annotated with information about what patterns indicate cancer and which do not.
AI analyzes images using deep learning models like convolutional neural networks (CNNs). CNNs use convolutional layers, which act as filters to show only certain parts of an image. These layers allow the AI to hone in on essential features of the image and disregard uninformative areas. The AI uses its “training” to calculate a risk score for potentially abnormal features. This score indicates the likelihood of malignancy.
Google’s AI software recently outperformed trained radiologists in detecting breast cancer from mammograms. The study, published in Nature, involved training the AI on mammogram datasets from over 28,000 women in the USA and the UK. The AI used three deep learning models to produce a score from zero to one, indicating the likelihood of cancer. The AI analysis led to significantly fewer false positives and negatives than the radiologists’. Notably, there were also instances where the radiologists correctly detected cancer that the AI missed.
Similar studies comparing AI and radiologists have been done since the introduction of more primitive image analysis software some twenty years ago. One study from 2021 compared the performance of an AI trained on over 288,000 breast ultrasound exams against ten certified breast radiologists. The test sample consisted of 44,755 ultrasound exams, and the AI was better at detecting cancer than the radiologists' average score. In a 2019 study, AI outperformed a majority of radiologists (61.4 percent of 101) in detecting cancer in a test set of 2,652 digital mammography exams.
AI can pick up on subtle differences in scan images, which are more challenging to notice with the naked eye, even for trained experts. This is particularly important for detecting early-stage breast tumors that are easier to treat and have better survival rates. AI can be rapidly trained on a massive number of annotated images, which is impossible for humans to match. Therefore, it can draw on more “experience” than any radiologist, allowing it to recognize a wider variety of patterns and abnormalities.
AI improves consistency in image analysis because it can execute precisely the same protocol each time. This provides an advantage over the global pool of radiologists, who may interpret the same image differently due to variations in their equipment and clinical experiences.
Another core advantage of AI is speed. AI can rapidly analyze thousands of images without the need for breaks or sleep. This means AI can quickly learn from previously validated scan images and derive diagnoses from new images in a fraction of the time it would take a human radiologist. Faster analysis benefits radiologists by allowing them to focus on more complex cases, make quicker decisions, and reduce their workload. Faster analysis also leads to more efficient diagnoses, which enables clinicians to focus on developing personalized treatment plans more quickly, prioritize urgent cases, and reduce patient wait times.
Early detection of cancer leads to better outcomes for patients. Screening programs have already improved breast cancer survival rates, and this is largely due to catching breast cancers at an early stage when they are more treatable. AI helps to find minute changes to breast tissue that can indicate the presence of cancer but are hard for radiologists to detect. Therefore, by integrating AI into breast cancer screening programs, there may be a higher chance that more cases will be caught early, ultimately leading to better survival rates for breast cancer.
Despite the exciting results produced by AI, several challenges and limitations must be considered when integrating AI into breast cancer diagnosis.
One potential risk is that we become over-reliant on AI, overlooking the benefits of human insights and intuition for diagnosis. While the studies we mentioned above showed that AI outperformed radiologists in many cases, it’s important to remember that the radiologists also outperformed the AI in some instances. These results clearly indicate that a balanced approach (often referred to as a hybrid approach) that takes insights from AI and human operators is optimal for accurate diagnosis. As it stands, overreliance, either on human insights or on AI-driven diagnosis will likely lead to less accurate results and ultimately suboptimal patient care.
The use of AI in health may contribute to existing racial inequalities in healthcare. For instance, black women in the US are more likely to have dense breasts than white women. This is significant because dense breast tissue is more challenging to analyze by mammography. Therefore, these populations are less likely to benefit from AI that has been trained to identify breast tumor cancers from mammograms. There are also concerns about turning over increasingly large amounts of medical information to a computer system that, no matter how well protected, will always be vulnerable to increasingly sophisticated hacking techniques, no matter how well protected.
AI requires datasets of annotated images to train itself on. If the available datasets are not representative of a wide population or are biased towards a particular subtype of breast cancer (of which there are many), then this limits the applicability of the AI. As we’ve seen from the studies described above, AI is still prone to false positives and negatives, which significantly impacts patient care.
So, when it comes to the question of whether Google AI (or any other AI, for that matter) beats the doctor at detecting breast cancer, the answer is yes, but that isn’t the whole story. AI is providing faster and more accurate diagnostics for breast cancer in many cases, but radiologists and doctors still play an essential role. For now, the evidence suggests a hybrid method where radiologists leverage AI to improve the accuracy and efficiency of their work is the best option. Future technological advancements, however powerful, must align with ethical principles to ensure optimal breast cancer detection and clinical decision-making.
If you want to be proactive about your health, why not book an Ezra full-body MRI? Our annual scan catches potential cancer earlier, leveraging AI through the screening process to make it more efficient, affordable, and faster.
