A0165 - Artificial Intelligence-Assisted Colon Polyp Detection: Initial Experience by Gastroenterology Fellows

Sunday, October 23, 2022

5:00 PM – 7:00 PM ET

Location: Crown Ballroom

Presenting Author(s)

Praneeth Kudaravalli, MD

Augusta University Medical Center
Augusta, GA

Praneeth Kudaravalli, MD¹, Haoran Peng, MD², Kwabena O. Adu-Gyamfi, MBChB³, Zain A. Sobani, MD⁴, Viveksandeep Chandrasekar, MBBS⁵, Humberto Sifuentes, MD², Kenneth J. Vega, MD, FACG⁵, John Erikson L. Yap, MD⁵
¹Augusta University Medical Center, Augusta, GA; ²Augusta University Medical College of Georgia, Augusta, GA; ³Medical College of Georgia - Augusta University, Augusta, GA; ⁴Augusta University/Medical College of Georgia, Augusta, GA; ⁵Augusta University Medical College of Georiga, Augusta, GA

Introduction: One-fourth of colorectal neoplasia are missed on screening colonoscopies. Polyp detection rate is an important quality indicator for colonoscopy performance. It varies widely among providers in both community and academic settings. Currently, professional societies recommend an adenoma detection rate of at least 20% for women and 30% for men. The aim of our study is to evaluate the use of artificial intelligence (AI) during colonoscopies performed by gastroenterology fellows to assist their detection of colon polyps.

Methods: Patients undergoing outpatient colonoscopy in two specific rooms equipped with GI Genius (Medtronic, Minneapolis, MN) from February to May 2022 were eligible for investigation. Inclusion criteria were patient procedures done by 2^nd and 3^rd year fellows supervised by one of 4 faculty members. Demographic data, colonoscopy quality measures and number of polyps were recorded. An inquiry by the faculty was performed with a binary response of yes or no for each detected polyp during the procedure to assess if AI assisted in fellow polyp detection. The polyp detection rate by AI was calculated as - (number of procedures AI assisted in detecting polyps / total number of procedures) x 100. In addition, data on if a change in endoscopic billing code occurred or surveillance interval was obtained as a result of AI. Descriptive statistics were used to analyze results.

Results: A total of 115 patients met inclusion criteria and comprise the study group. Average age of the participants was 56.9 yrs. Females were 57% and males were 43%. African Americans enrolled in accounted for 45.2%, Caucasians 50.4%, Hispanic/Latino 2.6% and 1.8% Asians. The mean Boston Bowel Prep Score was 7.5. Mean withdrawal time was 14.4 minutes. The AI assisted polyp detection rate was 46%. Reduced surveillance interval in 34.8% and increased procedure revenue in 32.2% occurred in patients with fellow performed AI assisted colonoscopy.

Discussion: The current investigation revealed that AI assisted colonoscopy resulted in an acceptable polyp detection rate, reduced post procedure surveillance intervals in those with AI detected lesions and increased overall revenue from trainee-performed procedures. AI improved quality of fellow colonoscopy exams by reducing missed lesions, allowing for correct patient stratification and maximal polyp/cancer reduction from the procedure. These preliminary results should be confirmed using larger cohorts of fellow procedures using AI during colonoscopy.

Figure: Image 1 – Graph on the top shows percent change in surveillance interval and the graph at the bottom shows percent change in diagnosis code.

Variable
Age	Mean	Std. Deviation
	56.9	11.9
Gender	N	%
Male	49	43
Female	65	57
Race	N	%
African American	52	45.2
Caucasian	58	50.4
Hispanic or Latino	3	2.6
Asian	2	1.7
BBPS Right Score	N	%
1	7	6.1
2	51	44.3
3	57	49.6
BBPS Transverse Score	N	%
1	2	1.7
2	48	41.7
3	65	56.5
BBPS Left Score	N	%
1	2	1.7
2	50	43.5
3	63	54.8
Total BBPS	Mean	Std. Deviation
	7.5	1.56
Withdrawal Time (minutes)	Mean	Std. Deviation
	14.4	10.19
Polyps Removed	Mean	Std. Deviation
Cecum
Total	0.21	0.58
AI Assisted	0.12	0.33
Ascending Colon
Total	0.44	0.9
AI Assisted	0.14	0.35
Transverse Colon
Total	0.53	0.99
AI Assisted	0.21	0.54
Descending Colon
Total	0.35	0.86
AI Assisted	0.1	0.33
Sigmoid Colon
Total	0.45	0.88
AI Assisted	0.12	0.37
Rectum
Total	0.19	1.02
AI Assisted	0	0
Entire Colon
Total	2.2	3
AI Assisted	0.7	0.9
Did this reduce surveillance interval	N	%
No	75	65.2
Yes	40	34.8
Did it change procedure code	N	%
No	78	67.8
Yes	37	32.2

Table: Demographics and Results.

Disclosures:

Praneeth Kudaravalli indicated no relevant financial relationships.

Haoran Peng indicated no relevant financial relationships.

Kwabena Adu-Gyamfi indicated no relevant financial relationships.

Zain Sobani indicated no relevant financial relationships.

Viveksandeep Chandrasekar indicated no relevant financial relationships.

Humberto Sifuentes indicated no relevant financial relationships.

Kenneth Vega indicated no relevant financial relationships.

John Erikson Yap indicated no relevant financial relationships.

Praneeth Kudaravalli, MD¹, Haoran Peng, MD², Kwabena O. Adu-Gyamfi, MBChB³, Zain A. Sobani, MD⁴, Viveksandeep Chandrasekar, MBBS⁵, Humberto Sifuentes, MD², Kenneth J. Vega, MD, FACG⁵, John Erikson L. Yap, MD⁵. A0165 - Artificial Intelligence-Assisted Colon Polyp Detection: Initial Experience by Gastroenterology Fellows, ACG 2022 Annual Scientific Meeting Abstracts. Charlotte, NC: American College of Gastroenterology.