AI Applications in Medical Imaging: Examples, Benefits, and Concerns

Artificial intelligence (AI) is a game-changing technology that uses computerized algorithms to decipher complex data. Diagnostic imaging is one of the most promising clinical uses of AI, and a lot of effort is going into establishing and fine-tuning its performance to make it easier to identify and quantify a wide range of clinical problems.

IT companies are capitalizing on this potential by creating artificial intelligence (AI) solutions to assist radiologists during diagnosis. These AI-powered medical image analysis technologies can detect early-stage illnesses, forecast the likelihood of cancer and neurological ailments, and allow clinicians to prescribe preventive therapies. Radiologists will save time and have more confidence in their choices with such instruments.

Examples of AI Applications in Medical Imaging

AI may be used in conjunction with a variety of imaging modalities to assist in the detection of anomalies in various areas of the human body.

Screening Breast Cancer

Research scientists from Google Health and DeepMind collaborated with Northwestern University’s Cancer Research Center and the Royal Surrey County Hospital to see how AI might perform in mammography cancer diagnosis. The researchers compared AI-generated mammography diagnoses to those made by human radiologists. This research included 26,000 women from the United Kingdom and 3,000 from the United States.

False positives were minimized by 5.7 percent among American participants and 1.2 percent among British women using AI. False negatives decreased by 9.4% in the United States and 2.7 percent in the United Kingdom. Some cancer indications that radiologists missed were identified by AI systems. Radiologists, on the other hand, discovered cancer indications that the algorithm missed. The findings showed that, for the time being, AI should be utilized as a supporting tool for radiologists and not as a viable alternative for human judgment.

Detecting Brain Tumors

Brain cancers require a long time to identify using traditional techniques. The diagnosis necessitates the use of a pathology lab, which entails a long delay for conventional processing, staining, and image analysis. Patients who are hospitalized for brain tumor surgery may not what kind of tumor they have, which adds to their worry.

The stimulated Raman histology (SRH) technology was developed by Michigan Medicine to create brain tissue pictures in less than 2.5 minutes, removing the normal processing stages. With the use of artificial intelligence, researchers were able to improve this approach even further.

Unveiling Neurological Abnormalities

In the early stages, various efforts are being implemented to identify illnesses like Alzheimer’s and Parkinson’s using AI. Newcastle University’s Octahedron project, in collaboration with Newcastle NHS Foundation Trust and Sunderland Eye Infirmary, is one such example.

The researchers are using retina OCT images to discover early symptoms of neurological diseases. The retina is the only component of the central nervous system visible and affected by illnesses like Alzheimer’s, without invasive treatments. High-quality OCT images are provided and are inexpensive to get. The AI algorithms were taught by the researchers to identify early indications of neurological conditions.

Spotting Fractures and Musculoskeletal Injuries

AI can improve the value that musculoskeletal imagers can provide their patients by enhancing image quality, patient centricity, imaging reliability, and diagnostic accuracy, as well as assessing the appropriateness of imaging orders and helping predict patients at risk for fracture.

Identifying Diabetic Retinopathy

Diabetic retinopathy (DR) is a prevalent diabetic condition and the primary reason for the avoidable blindness of individuals 20–74 years of age in the United States. Google and Verily collaborated to build an AI system for diagnosing diabetic retinopathy automatically. The AI system was evaluated by more than 2,000 patients in India at the Aravind Eye Hospital. Patients just have to gaze at the webcam with the AI algorithm.

Concerns about AI in Hospitals

Patient Data Privacy

Patients are regulated by the EU GDPR law, and researchers must obtain express authorization to the use of AI medical pictures by patients. In the case of patients’ consent, their data should be properly anonymised. Another intriguing issue is: are these patients eligible for financial reimbursement if they use patient data to advertise their AI products?

Another issue is data distortion. Bias occurs if the data used to train AI systems only represent one specific cohort and are not reflected correctly by the whole population. Minorities are frequently under-represented, for example, that means that radiologists need to look more closely at colors when they look at people who are mainly white.

Generalizability of AI Algorithms

Algorithms get their “intelligence” from training datasets. Even when a training set data is successfully used by software, the performance might decrease if it meets a fresh collection of data. In the course of training, firms and researchers in AI Medical Imagery need to remember where the algorithm is employed and which section of the population it serves.

If the algorithm was built in Europe and trained on local record-sets, it will be under-presented in the training package and not familiar with the algorithm in African nations, as will the frequent illnesses in Africa.

Radiologists Are Skeptical

The largest obstacle to implementing AI on a broader scale is the radiologist itself, according to a study performed by the McKinsey Company. The diagnostic capability of AI systems is not one hundred percent reliable. The absence of rules controlling AI deployment is a further concern on the part of the radiologists. For example, if there is a negative indication, who is responsible?

According to a study of European Society of Radiology members, the majority prefers that legal liability be divided among the parties concerned or placed solely on the shoulders of the radiologist.

How to Push AI in Medical Imaging

● Educate Radiologists about the Role of AI: The role customers want AI to play in their health facilities should be obvious to hospital administrators. During discussions with radiologists, they must stress that AI is merely another tool in the arsenal of the radiologist and not a substitute for its function. Even equipment that does not need physician intervention will not completely replace humans.

● Devise appropriate Methods for Data Sharing: The quality and variety of training datasets determine the performance of AI systems. Researchers and doctors can collaborate to create sets that are rich in annotations and information. If sharing datasets is not a possibility, temporary training access can be granted while the data remains safe behind the hospital’s firewall.

● Utilize inclusive Datasets for Algorithm Training: The data bias and generalizability problems described in the preceding sections should be considered by everyone responsible for generating training datasets. The metadata must reflect the demographic and environmental characteristics in the area where the algorithm will be used.

Bottom Line

Every stage in the imaging value chain, including interpretative and non-interpretive components, will be transformed by artificial intelligence (AI). To be leaders in clinical AI applications, radiologists should get conversant with AI development .

The healthcare system can benefit from AI-powered image processing in a variety of ways. Healthcare establishments, on the other hand, are not rushing to incorporate AI into their practices. It doesn’t help that several FDA-approved AI technologies have already failed to deliver on their promises.

Even though today’s AI technologies are of higher quality, everyone from doctors to firms that provide machine learning radiology solutions must contribute to their adoption.