According to a article made by a team of Google Artificial Intelligence researchers published in the scientific journal nature, a machine learning algorithm can detect an error in the chest X-ray.
The new machine learning system helps radiologists prioritize chest radiographs and can also serve as a first-response tool in emergency settings where radiology is not always available. Google announced that although this system is efficient, it will not, under any circumstances, serve to replace healthcare professionals. has the purpose increase productivity at a time when the world faces a severe shortage of experts.
This study demonstrates several advances when analyzing X-rays through Artificial Intelligence, however a medical analysis is essential to identify possible diseases. If the doctor does not know if the patient has a disease, and is not looking for a specific problem, the effectiveness of this system significantly decreases.
Most of the time radiologists are not necessarily looking for a specific disease. ÇBuilding a system that can detect all possible diseases is extremely difficult if not impossible, so this system is not a complement to the work of health professionals.
“The wide range of possible CXR abnormalities known as chest radiographs makes it impractical to detect all possible conditions by building several separate systems, each of which detects one or more pre-specified conditions,” Google researchers stress. “A reliable AI system that distinguishes normal from abnormal CXRs can contribute to the assessment and management of the patient’s situation,” they add, which is critical to setting priorities and making radiologists’ work more successful.
The Artificial Intelligence system was tested on more than 25 000 X-rays from five different hospitals in India, and Google researchers decided to provide “the abnormal versus normal labels from three radiologists (2430 labels on 810 images) for the test set CXR-14 publicly available” to make it easier to develop “AI models for chest radiography”.
Source: Venture Beat