|
AI diagnosis is more effective than human diagnosis in the diagnosis of pCLE imaging in the biliary duct cancer epithelium spreading
|
|
Koichi Furukawa1,
Naoyuki Yokoyama2,
Atushi Fujita3 |
|
1Div. Gastroenterology, Niigata City General Hospital, 2Div. Surgery, Niigata City General Hospital, 3KCCS Lit Co |
 |
|
| Background pCLE has the advantage of obtaining a magnification image that is like optical biopsy noninvasively, without the interference of bleeding and mucus secretion. However, it is sometimes difficult because the pCLE image is different from a pathological tissue image and unique signal. AIM Comparison AI diagnosis and human diagnosis of pCLE image of epithelium spreading of the biliary duct cancer Methods We made the classifier of extracted features of the cholangiocarcinoma pCEL images by deep learning ‘cafe’ base framework .They were compared to the pathologic examination of the surgical specimen and biopsy. After learning by 9 volunteer doctors with the same teacher images set used in the classifier, we compared AI and human diagnoses using common test images. Result AI classifier acquire accurate 69.8%. In the AI vs human diagnostic comparison of the test images, the sensitivity was 0.727 / 0.601, the specificity 1.0 / 0.654, the negative predictive value 1.0 / 0.597, and the positive predictive value 0. 667 / 0.67. All AI diagnoses had each kappa value of 1.0, confirming the reproducibility of diagnostic accuracy that can not be obtained by human diagnoses. Conclusion AI iagnosis is effective in diagnosis is effective in the pCLE image diagnosis of bile duct cancer with intraductal spread. |
Index Term 1: confocal laser endomicroscope
Index Term 2: computer-aided diagnosis
|
| Page Top |