imaging has more than tripled in the past 10 years, with the same number of
diagnosticians. That is, image diagnosis supported by machine learning and deep
learning is useful to realize efficient image diagnosis.
5.1 Pylori Infection Diagnosis with AI
According to Sotoki Shichijyo et al., AI for Helicobacter pylori infection diagnosis
exhibits better sensitivity and specificity than the average value of 23 doctors. A
22-layer, deep convolutional neural networks (CNN) was pretrained and fine-tuned
using a data set of 32,208 images with either positive or negative diagnosis for
H. pylori (first CNN). A separate test dataset (11,481 images from 397 patients) was
evaluated by the CNN and 23 endoscopists independently (Shichijo et al., 2017).
They concluded that gastritis caused by H. pylori could be better diagnosed on the
basis of endoscopic images compared with the manual diagnosis performed by
endoscopists.
In addition, the application of AI has been successful in diagnostic imaging of
cancer. A CNN-based diagnostic system was constructed by using a single-shot
multiBox detector architecture and trained using 13,584 endoscopic images of
gastric cancer. To evaluate the diagnostic accuracy, an independent test data set of
2296 stomach images collected from 69 consecutive patients with 77 gastric cancer
lesions was applied to the constructed CNN. The constructed CNN system for
detecting gastric cancer could process numerous stored endoscopic images in an
extremely short time with a clinically relevant diagnostic ability. It may be well
applicable to daily clinical practice to reduce the burden of endoscopists (Hirasawa,
Aoyama, et al., 2018).
5.2 Analyzing Endoscopic Images with AI
Colorectal cancer has been on the rise in recent years, ranking number 1 in the cancer
death rate in Japanese women and number 3 in men. As a countermeasure, it is
known that the mortality from colon cancer can be significantly reduced (53–68%)
by removing tumorous polyps that are early cancer and precancerous lesions with
colonoscopy (Zauber et al., 2012; Nishihara et al., 2014).
Among polyps, however, there are neoplastic polyps that need to be resected,
whereas non-tumor polyps (nonneoplastic polyps) that need not be resected. Physicians need to properly distinguish these polyps during time-constrained examinations. However, the judgment is not easy, as it is strongly dependent on the doctor’s
intuition and experience. Early detection and treatment of neoplastic polyps using an
endoscope have the effect of suppressing colon cancer death.
Under such circumstances, Susumu Kudoh et al. at the Digestive Center of
Yokohama City Northern Hospital in Showa University analyzed endoscopic
16
H. Matsushita
diagnosticians. That is, image diagnosis supported by machine learning and deep
learning is useful to realize efficient image diagnosis.
5.1 Pylori Infection Diagnosis with AI
According to Sotoki Shichijyo et al., AI for Helicobacter pylori infection diagnosis
exhibits better sensitivity and specificity than the average value of 23 doctors. A
22-layer, deep convolutional neural networks (CNN) was pretrained and fine-tuned
using a data set of 32,208 images with either positive or negative diagnosis for
H. pylori (first CNN). A separate test dataset (11,481 images from 397 patients) was
evaluated by the CNN and 23 endoscopists independently (Shichijo et al., 2017).
They concluded that gastritis caused by H. pylori could be better diagnosed on the
basis of endoscopic images compared with the manual diagnosis performed by
endoscopists.
In addition, the application of AI has been successful in diagnostic imaging of
cancer. A CNN-based diagnostic system was constructed by using a single-shot
multiBox detector architecture and trained using 13,584 endoscopic images of
gastric cancer. To evaluate the diagnostic accuracy, an independent test data set of
2296 stomach images collected from 69 consecutive patients with 77 gastric cancer
lesions was applied to the constructed CNN. The constructed CNN system for
detecting gastric cancer could process numerous stored endoscopic images in an
extremely short time with a clinically relevant diagnostic ability. It may be well
applicable to daily clinical practice to reduce the burden of endoscopists (Hirasawa,
Aoyama, et al., 2018).
5.2 Analyzing Endoscopic Images with AI
Colorectal cancer has been on the rise in recent years, ranking number 1 in the cancer
death rate in Japanese women and number 3 in men. As a countermeasure, it is
known that the mortality from colon cancer can be significantly reduced (53–68%)
by removing tumorous polyps that are early cancer and precancerous lesions with
colonoscopy (Zauber et al., 2012; Nishihara et al., 2014).
Among polyps, however, there are neoplastic polyps that need to be resected,
whereas non-tumor polyps (nonneoplastic polyps) that need not be resected. Physicians need to properly distinguish these polyps during time-constrained examinations. However, the judgment is not easy, as it is strongly dependent on the doctor’s
intuition and experience. Early detection and treatment of neoplastic polyps using an
endoscope have the effect of suppressing colon cancer death.
Under such circumstances, Susumu Kudoh et al. at the Digestive Center of
Yokohama City Northern Hospital in Showa University analyzed endoscopic
16
H. Matsushita