What does artificial intelligence do in medicine?

This article was written for The European Sting by our guest writer, Mr. Jakub Kufel medical student at Silesia Medical University, Poland. The opinions expressed within reflect only the writer’s views and not necessarily The European Sting’s position on the issue.

---

Artificial intelligence (AI) is a general concept that assumes the use of a computer to model intelligent behavior with the least human intervention. The term comes from the Czech word robot, which means biosynthetic machines used as forced labor. This term applies to a wide range of medical articles such as robotics, medical diagnosis, medical statistics, and human biology, up to today’s “omic”. Ai in medicine has two main categories: virtual and physical. Its virtual part deals with the analysis of AI from the management of deep learning of information to the control of health management systems, such as electronic health records or active advice of doctors in making treatment decisions. The physical aspect of AI relates to robots used to assist an elderly patient or operator in an operating room.

The use of artificial intelligence in medical imaging

AI particularly good achievements in this field are achieved in the analysis of brain scans by magnetic resonance imaging (MRI) to predict Alzheimer’s disease. Convolutional neural network (CNN) presented for the first time in the 1950s and gaining popularity over the last 10 years is successfully used in predicting the risk of knee osteoarthritis in MRI. AI is also gaining popularity in dermatology. In 2017, a team of researchers led by Esteva A. conducted a study in which artificial intelligence was found to have similar results in the diagnosis of skin biopsies compared to 21 licensed dermatologists. CNN has also been used for the early diagnosis of diabetic retinopathy based on fundus photos in high use. Radiology has the highest hopes for AI. Due to the large number of examinations performed and the small number of radiologists available, the evaluation of tests is delayed. There are already systems that use CNN to detect pulmonary tuberculosis or cancer on chest radiographs and computed tomography. To support the development of deep learning as part of artificial intelligence and encourage work on new solutions, the National Institutes of Health (NIH) has created and shared a database of over 100,000 chest radiographs. Several research teams have created a model using different neural networks to achieve high results in the disease entities visible on X-ray images. The teams subsequently achieved the following results: Wang et al. 75%, Shen et al. 78%, Guendel et al. 81%, Yan et al. 83%, Baltruschat et al. 73% (percentages are the average of correct diagnoses obtained from a shared analysis by NIH chest radiographs). This proves that CNN can be used in medical practice because they achieve similar results to those obtained by radiologists. In this way, they can relieve radiologists, support diagnostics in the future and, after improving, they can even replace doctors.

Why AI is needed in medicine?

AI is a phenomenon that changes the face of medicine. It is a field of science that is constantly evolving. CNN is used in almost every type of medical imaging. Particularly applicable in radiology. The constantly evolving AI will be entering new fields and medical specializations because it has not yet been applied in all of them. Further research is needed on the procedures implemented with the use of AI, as the current knowledge on this subject is relatively small. The implementation of appropriate technological and legal solutions will relieve health care and make access to it relatively simpler. References

1. Hamet P. i Tremblay J., Artificial intelligence in medicine, Metabolism: clinical and experimental, (2017), S36-S40.
2. Wang C., Elazab A., Wu J. i Hu Q., Lung nodule classification using deep feature fusion in chest radiography, Computerized medical imaging and graphics the official journal of the Computerized Medical Imaging Society, (2017), s. 10–18.
3. Shen Y. i Gao M., Dynamic Routing on Deep Neural Network for Thoracic Disease Classification and Sensitive Area Localization, MLMI, (2018).
4. Charte F., Rivera A., del Jesus M. J. i Herrera F., A First Approach to Deal with Imbalance in Multi-label Datasets, w: E. Hoz, A. Ortiz, J. Ortega, E. Hoz (red.), Network Anomaly Classification by Support Vector Classifiers Ensemble and Non-linear Projection Techniques, t. 8073, Berlin, Heidelberg, Springer Berlin Heidelberg, (2013), s. 150–160.
5. Baltruschat I. M., Nickisch H., Grass M., Knopp T. i Saalbach A., Comparison of Deep Learning Approaches for Multi-Label Chest X-Ray Classification, Scientific reports, (1), (2019), s. 6381.