Google AI assists breast cancer screening misdiagnosis rate is lower than human!

At the start of the new year, Google Health brought us good news, and its latest artificial intelligence model has a new breakthrough in breast cancer screening! Breast cancer affects women around the world, with more than 55,000 people diagnosed with the disease in the UK each year, and about one in eight women in the US at risk of developing breast cancer in their lifetimeIn countries with high incidence of breast cancer and advanced medical technology, such as the United States and Britain, although digital mammography technology is widely used, it still faces considerable challenges in detecting and diagnosing diseasesGoogle has worked closely with DeepMind, Cancer Research UK, Northwestern University and royal Surrey County Hospital to see if AI can help radiologists more accurately detect signs of breast cancer and find the best way to improve breast cancer screening techniquesThe team published its preliminary results in the journal Naturethey found that the AI model was able to correctly screen for signs of breast cancer in screening images at a level similar to that of experts, reducing the results of false negatives (also known as missed rates) and false positives (also known as misdiagnosis rates)doctor's misreading of the tablets can lead to false negative or false positive reports for breast cancer patients, such misdiagnosis will not only lead to patients do not get timely and effective treatment, but also bring unnecessary psychological pressure, and give radiologists more workUsing artificial intelligence technology, it is possible to improve the occurrence of these problemsthe AI system consists of three deep learning models, each of which is used for different levels of analysis, for the analysis of three situations: single lesions analysis, single breast analysis, and overall case analysisEach model produces a cancer risk score between 0 and 1 for the pathology of the mammogram, and the predictive accuracy of the overall AI system is combined with the average of the three modelstwo large data sets used to train AI models come from the UK and the USThe UK data set was collected from three mammography screening sites of the NHS Breast Screening Programme (NHSBSP) and contained mammograms of more than 76,000 women, while in the US it collected mammograms of about 15,000 patients at North West Memorial Hospital in Chicago between 2001 and 2018then used a separate data set to assess the reliability of the training, which consisted of more than 25,000 patient images (25,000 in the UK and 3,000 in the United States)After the evaluation was verified, the model's prediction results were both reduced by error, with AI reducing false positive situ and 9.4% false negative report by 5.7% compared to clinical practice data in the United States, and AI reducing false positive report and 2.7% false negative report by 1.2% compared to clinical practice data in the United KingdomThis is a "leap forward" for the current 20 per cent breast cancer screening rateBreast Cancer Prediction Performance: Artificial Intelligence Systems vs Clinicians (Photo: Resources 1)in addition, the researchers verified whether the AI model could be used in other medical systemsFirst, they trained AI based only on data sets from patients in the UK, and then used the data set of American patients as an evaluation and validation set for the model, in a separate experiment in which AI not only had higher accuracy in breast cancer prediction than human experts, while the reported rates of false and negative negatives were reduced by 3.5% and 8.1%, respectively, which meant that the AI system had the potential to be applied to other medical environments finally, to further verify the reliability of the system, the researchers called on six radiologists to "read the figure PK" with AI, which improved the accuracy of AI identification by explaining 500 cases from patients in the United States at the same time, according to the researchers, the system is essentially designed to better assist doctors rather than replace them, so the team's current results are enough to prove that human and AI doctors are complementary in future medical environments AI reduces the rate of missed detection, and human doctors can point out cases of disease syilation that AI has not identified no six experts identified the sample cases, but AI successfully identified sample cases, (b) 6 sample cases that were successfully identified by experts, but ai omitted sample cases (Image Source: Resources Source: 1) it is also worth noting that AI can still identify breast cancer more accurately with less historical information Compared to the human doctor's decision-making when there is a patient's medical history and historical data traceable, AI only deals with the identification of its immediate x-ray image, there is no historical information for reference, from this point, AI decision-making process is "more independent." looking to the future, the initial results of the study are promising, the model will help improve the accuracy and efficiency of breast cancer screening procedures, while also reducing the time and pressure of patients waiting for diagnosis results However, clinical medicine is complex, and the doctor's decision is not a simple binary decision (between the presence or not of cancer), but also must consider other signs and symptoms Obtaining other data is also technically complex, so in the future it might be better to use systems that can query electronic medical records to identify and annotate specific cases in conjunction with the latest AI References: McKinney, S.M., Siennek, M., Godbole, V et al International evaluation of an AI system for breast cancer screening Nature 577, 89-94 (2020) doi:10.1038/s41586-019-1799-6 s https:// Retrieved Jan 2, 2020 from https://venturebeat.com/2020/01/01/google-healths-ai-identifies-breast-cancer-in-mammogram-imagery-with-fewer-false-positives/ 4 s ingle aI to improve breast cancer screening Screening Photo2, 2020 from https:// 5' Artificial intelligence can help breast screening save more lives lives Jan 2, 2020 from https://scienceblog.cancerresearchuk.org/2020/01/01/artificial-intelligence-could-help-breast-screening-save-more-lives/ 6 s i show promise for the breast cancer screening screening Jan 2, 2020 https://