How Does AI Help with Disease Prediction, Diagnosis and Chronic Disease Management?

In recent years, the application of artificial intelligence in medical fields has been extensively broadened, covering different scenarios such as medical diagnosis, disease prediction, image analysis (radiology, histology), text recognition and natural language processing, drug activity design and gene mutation expression prediction, health management, treatment effect and prognosis prediction, as well as the rapid development of omics technology. In 2000, the U.S. Food and Drug Administration (FDA) approved the Da Vinci Surgical System produced by the American company Intuitive Surgical to be marketed. This minimally invasive surgery system can be used for complex operations such as urology, heart valve repair, and gynecology. More than 5,000 units of this system are currently in operation all over the world. In addition, a unique new type of precise drug delivery nanorobot has also been widely used.

Artificial intelligence has improved learning capabilities and is changing the future of healthcare. We’ll explore how AI will influence disease prediction, diagnosis and chronic disease management.

Disease prediction, diagnosis and chronic disease management
The application of AI can help perform the early diagnosis of diseases in many cases. The application research of AI in medical and health mainly focuses on cancer, nervous system and cardiovascular diseases, because these diseases are the main causes of disability and death. Contagious and chronic diseases (such as type II diabetes, inflammatory bowel disease, Clostridium difficile infection, etc.) have also received considerable attention. For example, the US Food and Drug Administration (FDA) allows the application of diagnostic software to detect wrist fractures in adult patients. In another study of 1,634 images of cancerous and healthy lung tissue, the algorithm successfully and accurately distinguished two common types of lung cancer from healthy conditions. The same work was done by three pathologists. Through image heat map pattern recognition, the accuracy of predicting major depression has reached 74%. Although noisy data and experimental limitations reduce the clinical utility of the model, deep learning methods can solve these limitations by reducing the dimensionality of the data through hierarchical automatic coding analysis. In an experiment with more than 1,400 images, both AI and experts are required to analyze 308 skin histopathology to detect basal cell carcinoma and distinguish benign and malignant lesions, and the diagnostic accuracy rate of AI reached more than 90% compared with experts.

In brief, the image recognition technology using artificial intelligence systems is able to interpret patients’ images, identify key information, give preliminary diagnosis, and help radiologists improve the diagnosis efficiency.

Gene chips are widely used to detect gene expression in cancer cells. However, although the chip has 20,000-50000 diagnostic probes with genetic features, noisy data and experimental limitations reduce their clinical application. Deep learning solves this problem by reducing data diversity (dimensionality), applying hierarchical automatic coding analysis, and training artificial neural networks to achieve more accurate cancer detection and classification. For example, using the deep learning architecture visual model to analyze the histopathology of 1,417 skin images, detecting basal cell carcinoma and distinguishing malignant and benign lesions, is superior to previous automated analysis, and the diagnostic accuracy rate is >90% compared with that of experts. Deep learning histopathology to identify metastatic breast cancer in non-lymph node biopsy has similar diagnoses to that of experts. These systems are trained by comparing the characteristics of millions of tumor-positive and negative histological plaques, using heat maps to post-process the data to predict the probability of tumors. When pathologists and deep learning are combined to optimize performance, the human error rate can be reduced by 85%.

Artificial intelligence technology has been applied to cardiovascular medicine to explore new genotypes and phenotypes of existing diseases, improve the quality of patient care, increase cost-effectiveness, and reduce hospital readmission and mortality. In the past decade, some machine learning techniques have been used for the diagnosis and prediction of cardiovascular diseases. In the near future, AI will lead to a paradigm shift to precision cardiovascular medicine.

AI can also optimize the care trajectory of patients with chronic diseases, provide precise treatment recommendations for complex diseases, and reduce medical errors. Artificial intelligence analysis can be used for chronic disease management. For example, retinopathy can be predicted by machine learning. Deep learning is used to detect and grade diabetic retinopathy and macular edema. After the ophthalmologist grades each image 3 to 7 times, it has high specificity and sensitivity for detecting moderately severe retinopathy and macular edema.