Recently, a team led by clinicians at Beth Israel Deaconess Medical Center and Harvard Medical School demonstrated that an artificial intelligence (AI)-based computer vision system can enhance screening accuracy of colon cancer. Tyler M Berzin, a gastroenterologist from Beth Israel Deaconess Medical Center, discusses how AI-based computer-vision algorithms can assist physicians. Let us examine how this is accomplished.
How does AI review images in real-time?
According to Tyler, this would be a real-time application of artificial intelligence, which is also rather unique. In clinical medicine, the majority of examples of AI applications occur after the initial patient engagement, for example, during the subsequent evaluation of the X-ray. However, the researchers require immediate assistance throughout colonoscopy monitoring, when the physician’s role is to methodically visually inspect the whole colon lining in order to find and dislodge precancerous polyps.
Among the colonoscopy and the endoscopy monitor, the suggested AI technology analyses the colonoscopy image. The screen shows the actual colonoscopy operation, but with blue or green warning boxes indicating suspected polyp locations. Thus, this is the ideal example of AI improving, not replacing, medical performance.
AIM Daily XO
Join our editors every weekday evening as they steer you through the most significant news of the day, introduce you to fresh perspectives, and provide unexpected moments of joy
Your newsletter subscriptions are subject to AIM Privacy Policy and Terms and Conditions.
How AI recognises Polyps images
In this scenario, the approach is founded on a deep-learning computer vision system, which is designed to learn how to recognise specific objects, provided enough samples of their appearance. It must be fed a large amount of visual data. Then, following a period of training, it acquires how to recognise the polyps. What’s intriguing is that these AI-based deep-learning algorithms may uncover traits that a physician would miss entirely. There are numerous examples of this, such as a deep-learning algorithm for X-rays.
Detection Improvement
The research reveals that doctors were approximately 30% less likely to ignore a polyp when assisted by AI. Independent, external evaluation of AI medical algorithms is a critical aim for AI in medical practice. This will be the first projected randomised trial to externally evaluate an AI algorithm’s efficiency.
Download our Mobile App
Other AI Research Contributions
- In 2015, a research team from Italy used artificial neural networks to examine the relationship between hereditary and environmental variables and DNA methylation in colorectal cancer (CRC).
- After a few years, Chinese researchers used data from The Cancer Genome Atlas (TCGA) database and machine learning techniques to enhance CRC screening.
Dataset: The Cancer Genome Atlas (TCGA) database
- Additionally, the researchers from the USA proposed a machine learning technique that employed tumour-derived cell-free DNA to achieve excellent sensitivity and specificity. Their machine learning technique may provide an exciting new route for future research in early-stage colorectal cancer detection.
- A group of researchers from Ireland compared 380 microRNAs (miRNAs) expression profiles in stage II colorectal tumours and normal tissues using artificial neural networks.
- A team of researchers from Iran developed a unique computational technique employing a Naive Bayes classifier to optimise this method and increase prediction accuracy.
- Another study conducted by a Spanish research group investigated the effectiveness of plasma samples when used in conjunction with a powerful predictive model for the distinction of healthy individuals from patients with colorectal cancer and advanced adenomas. The SVM classification model achieved a sensitivity of 85% and a specificity of 90%.
- A group of researchers from Iran and the USA suggested an artificial neural network model that effectively identified sample data as malignant or non-cancerous from the Gene Expression Omnibus (GEO) database.
Dataset: Gene Expression Omnibus (GEO) database
- Additionally, the Chinese researchers recommended the use of a dual CNN prediction model to identify potential disease-associated miRNAs. Their AI-based method investigates the deep commonalities between miRNAs, illness similarities, and miRNA-disease connections.
Conclusion
Currently, AI technologies allow clinicians to forecast the future of patients. In general, integrating AI technologies into the screening, diagnosis, and treatment of colorectal cancer may enhance patients’ clinical results and prognoses. Deep learning algorithms have been increasingly employed in clinical cancer research in recent years. Numerous modern artificial intelligence systems have demonstrated promising results in accurately detecting and characterising suspected lesions. However, more prospective, large-scale, multicenter clinical trials are needed to assess AI systems‘ diagnostic accuracy.
