The next big think: Artificial intelligence arrives at Penn Gastroenterology
Penn Gastroenterology and Hepatology has introduced GI Genius™, the first FDA-approved device to use AI for colon polyp detection during colonoscopy.
In 2022, Penn Gastroenterology and Hepatology received a grant for the department to incorporate GI Genius™, the first FDA-approved device to use artificial intelligence (AI), to assist clinicians in colon polyp detection in real time during a colonoscopy.
“This specific grant was to offer AI technology for underserved communities,” says Penn gastroenterologist Neilanjan Nandi, MD, FACP. “We applied for this grant in order to enhance colon polyp detection rates in our West Philadelphia community with the ultimate goal to further reduce colon cancer in our local population.”
Background:
- AI is the study of algorithms that give machines the ability to reason and perform cognitive functions, and encompasses the field of machine learning and its inter-related sub-domains.
- Colorectal cancer (CRC) is the third leading cause of death from cancer in the United States. Preventative screening for colorectal cancer can significantly reduce disease risk and mortality. A collection of challenges, including lack of access to care and irregularities in insurance coverage, have historically presented barriers to screening in medically underserved populations.
Improving access to care in the underserved community is among the founding missions of Penn Medicine. With regard to colorectal cancer in particular, this means access to colonoscopy and other essential screening modalities, and participation in screening programs. Cohort studies in targeted populations suggest that colonoscopy can bring about substantial reduction in colorectal cancer mortality.
Periodic screening and surveillance colonoscopy can substantially reduce the risk of colorectal cancer development and reduce associated mortality by catching polyps early.
However, despite regular lifetime adherence to a colon polyp surveillance schedule, there is still a 5 to 9 percent risk of developing an “interval” colon cancer. This inherent small risk is due to the potential to miss small polyps during a colonoscopy that may evade detection and transform over time into a cancer. High quality preparation and high-quality endoscopic skill are necessary to maximize the adenoma detection rate (ADR).
The national ADR benchmark is 26 percent across the country. Penn Medicine gastroenterologists currently significantly surpass this benchmark at an approximate 38 percent ADR.
While early detection of stage 1 to 3 colon cancer is associated with at least 85 percent five-year remission rates, optimizing ADR to find and resect adenomas is the key to complete colon cancer prevention. Therefore, tools that can aid in the increased visualization of colon adenomas can help the endoscopist remove these polyps to ultimately prevent future colon cancer pathogenesis.
“The aim of GI Genius™ is to apply AI technology during colonoscopy in order to detect more colon polyps and reduce adenoma miss rates so that people don’t develop interval colon cancers,” explains Dr. Nandi.
GI Genius™: Endoscopy powered by AI
GI Genius™ is composed of an artificial intelligence algorithm that aids endoscopists in highlighting portions of the colon where a potential polyp may be lurking.
During colonoscopy, the system superimposes real time markers on the endoscopic camera video feed that suggest the presence of a potential lesion being identified. These signal to the clinician that further assessment and intervention may be needed. In real-time, the endoscopist may need to provide closer visual inspection, tissue sampling, and/or polyp resection or ablation.
Ultimately, it’s up to the clinician’s clinical discretion on how to manage a colon polyp per their best clinical judgement and per standardized clinical practice guidelines.
The safety and effectiveness of GI Genius™ was studied through a multicenter, prospective, randomized, controlled study with 700 individuals ranging in age from 40 to 80 years old. These individuals were undergoing colonoscopy for colorectal cancer screening, colon polyp surveillance, positive fecal immunochemical test for blood in the stool, or reported gastrointestinal symptoms of possible colorectal cancer.
Study subjects underwent either standard white light colonoscopy alone or standard white light colonoscopy with the GI Genius™. Study results illustrated that colonoscopy utilizing GI Genius™ was able to identify pathology-confirmed adenomas or carcinomas in 55.1 percent of patients compared to 42.0 percent of patients with standard colonoscopy, an observed difference of 13 percent.
“We are still in the early days of adopting AI in colon cancer prevention as routine, but the data suggests that it can improve ADR and it may become a standard of care in the years to come,” says Dr. Nandi.
Remaining challenges include the additional financial costs of new technology and insurance reimbursement. In the interim, Penn Presbyterian Gastroenterology is proud to pilot this technology at no cost to our community.
Studying Artificial Intelligence at Penn
The study of artificial intelligence is being advanced throughout Penn Medicine at both the Perelman School of Medicine and the Abramson Cancer Center.
Recent developments include:
- AI to uncover hospital patients’ long COVID care needs -School of Medicine researchers used a new machine learning technique to help uncover hospital patients’ long COVID care needs—an approach that could be used more broadly to inform clinical decision making in hospital systems.
- Electroencephalography creating an EEG database: A comprehensive clinical electroencephalographic resource from four Boston hospitals - Nishant Sinha, PhD, of the Perelman School of Medicine is contributing to the Harvard Electroencephalography Database--a large-scale standardized electroencephalographic (EEG) resource supporting artificial intelligence-driven and reproducible research in epilepsy and broader clinical neuroscience.
- Novel machine learning to repurpose FDA-approved drugs Translational medicine expert David Fajgenbaum, MD, and colleagues are using a range of scientific approaches in clinical research—including an AI-guided prediction system to analyze FDA-approved medications that could be repurposed as potential treatments for patients with rare diseases.
- MRI-based AI model for the identification of underlying atrial fibrillation after ischemic stroke: A multicenter proof-of-concept analysis - Saman Nazarian, MD, PhD, is participating in an international novel end-to-end AI model that uses MRI data to rapidly identify high AF risk in patients who suffer from an acute ischemic stroke.
- Clinical correlates of two neuroanatomical AI dimensions in the Alzheimer’s disease continuum - Artificial Intelligence in Biomedical Imaging Laboratory Center for AI and Data Science for Integrated Diagnostics (AI2D) at the Perelman School of Medicine is applying artificial intelligence to deepen our understanding of the multifaceted pathogenesis of Alzheimer’s disease beyond the brain.
- “AI scribe” technology means more “face time” with patients - In a JAMA Network Open study, Penn researchers detailed how an AI ‘scribe’ increases face-to-face time with patients – and is associated with greater clinician efficiency, lower mental burden of documentation, and a greater sense of patient engagement during outpatient visits.
- Personalized left ventricular hypertrophy thresholds for hypertrophic cardiomyopathy diagnosis - Cardiologist Anjali Tiku Owens, MD, and Radiology, Associate Vice Chair of Research IT Walter Witshey, PhD, were participants in a study that used cardiovascular magnetic resonance and a validated artificial intelligence algorithm to measure left ventricular maximum wall thickness (MWT) to determine that the current >15 mm threshold for MWT in left ventricular hypertrophy ignores demographic influences.
- Readability of AI and ophthalmologist responses to patient surgery queries - Researcher John C. Lin, MD, et al, found that Chat GPT-4, the Large Language Model that understands and generates human-like text for chat bots, used a higher percentage of complex words compared to ophthalmologists, in communications with patients.
Clinical consult and patient referral
To speak with a provider or to refer a patient to Dr. Nandi or another provider at Penn Medicine, please call 877-937-7366, or submit a referral through our secure online referral form.