Google AI diagnoses lung cancer more successfully than radiologists

Google AI diagnoses lung cancer more successfully than radiologists

We are searching data for your request:

Forums and discussions:
Manuals and reference books:
Data from registers:
Wait the end of the search in all databases.
Upon completion, a link will appear to access the found materials.

How Reliable Is Artificial Intelligence For Diagnosing Cancer?

Google has now introduced a new artificial intelligence (AI) system that can detect lung cancer in early tests and even exceeds the diagnosis of radiologists in its accuracy.

Google AI's current research has developed artificial intelligence that can be used to diagnose lung cancer. The accuracy of the diagnosis of AI even surpasses human radiologists. The results of the investigation were published in the English-language journal "Nature Medicine".

AI was more reliable than human radiologists

Technology giant Google trained the algorithm on 42,000 CT scans of patients who had participated in a clinical study by the National Institutes of Health. The diagnoses of cancer created thereby exceeded the diagnostic accuracy of six radiologists involved. The AI ​​found five percent more cancers and 11 percent fewer false positives when evaluating a single scan, meaning cancer was suspected but the knot was actually harmless.

Improved precision is a major advance

The new technology from Google is able to improve the precision of the screening enormously, the researchers report. Even skeptics of lung cancer screening said the results said Google's performance in reducing misdiagnosis was a significant advance. The new AI could prevent people from being negatively influenced in their lives by an incorrect diagnosis and then being treated incorrectly.

The AI ​​must undergo further tests

So-called Artificial Intelligence (AI) has long been seen as a way to improve the screening of diseases and thus enable a number of malignant tumors to be localized with far greater accuracy. However, the Google system has to undergo even more rigorous testing before it can finally be implemented in medical practice. Since the study was limited to patients who had already been treated, it is not yet known whether the system would also lead to similarly good results in new patients.

Accuracy was 94.4 percent

Like several other so-called deep learning algorithms, which have already been tested for use in medical fields, the AI ​​was trained in this study using scans from previous lung cancer screenings. Then the artificial intelligence was commissioned to assess over 6,700 cancer screenings. This was to determine what accuracy can be achieved in diagnosing cancer. The presence of cancer in these cases was already known to the doctors. In the test carried out, the AI ​​achieved an impressive accuracy of 94.4 percent, report the authors of the study.

Cooperation between AI and doctors optimizes the treatment

The algorithm's statements were then compared to the diagnosis of six human radiologists to determine how well AI and humans could diagnose cancer using scans they had never seen before. If additional information in the form of tomography scans was available for screening data sets, the performance of the AI ​​was comparable to that of human experts. In cases where no additional tomography data was available, the AI ​​was able to clearly outperform its human colleagues with eleven percent less positive misdiagnoses and five percent less false negative diagnoses. This may sound like bad news to human doctors and medical professionals at first, but it's not about replacing professionals with an AI. Doctors will continue to play an important role in diagnosing and treating cancer. However, by combining the experience of a human doctor with the mind of a deep learning algorithm, it is likely that fewer mistakes will be made in the diagnosis, which would lead to an overall better quality of life for all concerned. (as)

Author and source information

Video: Keynote Fireside Chat: Cloud and AI in Healthcare and Biomedical Research Cloud Next 18 (November 2022).