How AI Can Help Streamline the Patient Experience

Years ago, a loved one was referred for a breast mass biopsy. She watched as the clinician glanced at her two-dimensional mammograms, palpated the region, and proceeded to take a sample of what fortunately turned out to be the correct tissue. A tracking tag was placed in the biopsy location to allow comparisons on future mammograms. The sample tissue was later confirmed as benign.

Other friends hadn't been so lucky. Return visits were needed for repeat biopsies as a result of insufficient samples being taken. Each subsequent repeat biopsy required hours on the phone during working hours with the hospital scheduling department in hopes of securing an appointment, discussions with insurance providers to convince them a repeat biopsy was required, another hour waiting for the procedure room to become available after arriving 15 minutes early (or else the procedure would be canceled), and yet another day off work. On top of this was enduring more anxiety waiting for the procedure and pathology results.

Such a scenario is the norm, even in a major metropolitan area with what's considered easy access to primary and specialist caregivers. Nowadays, a different experience is on the horizon, thanks to advances in artificial intelligence (AI).

Where AI Can Help

At the beginning of this article series, Pat Baird discussed how AI's promise lies in its capability to augment or free up humans to perform routine tasks better, faster, or more consistently.¹ But what about its potential to free up patients from painful, duplicative, tedious, or poorly managed healthcare interactions? These interactions consume hours, dollars, and goodwill and place barriers to care that are unsurmountable to some and inconvenient to all. Let's walk through a patient's journey to see where AI can help.

Before the Patient Encounter

Our patient's journey starts prior to screening and diagnosis. Getting patients to show up for screening can be a challenge in itself, and researchers in Singapore have developed an AI-based model to predict which patients might be no-shows for outpatient magnetic resonance imaging (MRI) appointments.² These patients received reminder telephone calls, reducing no-show rates dramatically among patients receiving calls.

Once the patient arrives, AI can help make sure their caregivers are ready for the encounter. The American Academy of Family Physicians' Innovation Lab is working with an AI-enabled patient management software application to help clinicians prepare for and navigate patient encounters.³

Enhancing the Procedure

And now we get to the screening procedure itself. Although the conventional mammography remains the standard of care for screening, breast ultrasound is increasingly used as a supplement to mammography in screening and as a primary modality in diagnosis, especially when the patient has dense breast tissue. However, adding ultrasound to mammography screening has been associated with high false-positive rates and more biopsies without necessarily catching more cancers.

AI has the potential to benefit patients and clinicians in a variety of ways. In the case of mammography screening, AI can help, for example, lower false-positive rates, reduce the need for further imaging and repeat visits by making sure images are acquired correctly the first time, improve patient positioning during CT and MRI scans, and prioritize the most critical scans so they are viewed by clinicians more quickly.

In 2021, a team of researchers, data scientists, and engineers from New York University published results from the use of an AI system that provided overreads of breast ultrasound B-mode imaging and color Doppler data.⁴ The authors focused their model's design on both explainability (i.e., being able to describe to the user why a lesion was suspicious) and the ability to reduce false-positives. The model was trained on millions of images from almost 300,000 screening and diagnostic ultrasound exams. The AI model helped radiologists decrease their false positive rates by 37% and reduce requested biopsies by 28% without affecting the overall sensitivity for cancer detection.

These systems are entering clinical use now, with suppliers such as Koios and Samsung marketing AI-enabled software for tumor characterization for both breast and thyroid ultrasound applications. Consider the wasted dollars, hours, and stress on each and every patient, caregiver, and facility involved in a false-positive referral for biopsy—and what better uses we could put them to.

If our patient is referred for further imaging, AI is here to help make sure their images are acquired correctly and on the first try. Growing up, I always dreaded getting dental X-rays for fear of needing to sneeze, wiggle, or cough in the 15 seconds I was required to sit oh-so-still. Now, camera-based systems using AI-enabled image processing are available from several CT vendors, including GE, Philips, and Siemens, to enhance patient positioning and optimize imaging every time.

These systems use anatomical landmarks that are properly captured with less manual input from a technologist. This reduces the time required to set up and position the patient, ultimately benefiting the facility by allowing a few more exams to be performed each day. The technologist also benefits from not having to manually readjust the patient, while the patient gets to spend less time lying still on a hard table.

In addition to patient positioning, AI is improving the scan itself. In the case of computed tomography, several diagnostic imaging vendors offer AI-enabled software to reduce and optimize radiation dose, lower image noise, and improve detectability of low-contrast lesions. A potential additional benefit is that the patient doesn't receive excess radiation, which can itself raise cancer risk over time.

Other modalities also are benefiting from AI-enabled tools. In MRI, AI-aided positioning helps reduce the time the patient spends in the MRI room, which can be particularly important for anxious patients. Likewise, AI-enabled reconstruction algorithms can allow high-quality images to be produced from less data, allowing for a shorter MR exam duration (e.g., 45% shorter in threedimensional brain imaging⁵) and increasing both patient throughput and patient comfort.

Prioritization and Access

After the scan is acquired, AI is here to help ensure the most critical scans get viewed first. Patient-prioritization software packages identify potential abnormalities in acquired images and move them higher in the clinician's queue so that they are read more quickly. This has several benefits for radiologists, who can focus more on complex cases, and for patients, whose critical conditions are diagnosed more quickly. AI can also help patients who live in remote geographies with limited access to specialists get the care they need closer to home by supporting teleradiology.⁶

Conclusion

Although the mental burden of a potentially life-threatening diagnosis will never get lighter, I'm hoping that with AI's help, we can make the journey less burdensome and a little bit more humanized for patients and clinicians alike. That is, as long as we make sure we've properly selected our training and validation data sets, of course.⁷