Sat, Oct 31, 2020 page2

• NTU Hospital researchers develop AI system to help spot pancreatic cancer

  • Staff writer, with CNA
  • • 
    • 
    • 

  Researchers at National Taiwan University Hospital (NTUH) have developed an artificial intelligence (AI) system that can identify tumors in the pancreas with an accuracy of better than 90 percent.

  NTUH doctor Liao Wei-chih (廖偉智) on Tuesday said that pancreatic cancer was the seventh-deadliest type of cancer in Taiwan last year, accounting for nearly 2,500 deaths.

  The disease is extremely hard to detect, as people show no symptoms in the early stages, Liao said.

  Studies have found that 40 percent of pancreatic tumors smaller than 2cm are missed in computed tomography scans, he said.

  This is because small tumors are not lumps, but appear to be a thin gray film, which is a challenge for even the most experienced expert to identify, he said.

  As a result, people are often only diagnosed when the cancer has spread to other parts of the body, which complicates treatment, he said.

  To determine whether AI could help detect tumors and help radiologists interpret CT scans, Liao and his team used convolutional neural networks (CNNs) to analyze images.

  The deep-learning system was introduced in a study called “Deep Learning for Pancreatic Cancer Detection: Current challenges and future strategies” published in the June edition of The Lancet Digital Health.

  The researchers used CT scans from 295 people with pancreatic cancer and 256 control images from people without cancer to train the AI system to distinguish between the two.

  They tested the accuracy of the system against CT images from 176 people with pancreatic cancer and compared the results with how radiologists judged them.

  The study found that “CNN can accurately differentiate pancreatic cancer from non-cancerous pancreas... CNN holds promise for developing computer-aided detection and diagnosis tools for pancreatic cancer to supplement radiologist interpretation.”

  The system detected 35 of 38 tumors smaller than 2cm (92 percent), 92 of 92 tumors between 2cm and 4cm, and 46 of 46 tumors 4cm or larger, the study said.

  That compared favorably with NTUH radiologist detection rates of 89.5 percent for tumors under 2cm (34 of 38), and 90.8 percent for tumors between 2cm and 4cm (79 of 87).

  The study warned against reading too much into the higher sensitivity achieved by the CNN-driven software than the radiologists, as the program examined whether pancreatic cancer existed in areas chosen by radiologists for examination.

  “Collectively, our results suggest that the CNN could supplement radiologists to reduce the miss rates rather than outperform or replace radiologists,” the study said.

  That data were based on images from NTUH, but tests were also done using images from the US, where the results were not quite as good, it said.

  The US system’s sensitivity was 63.1 percent for tumors under 2cm (41 of 65), 82.3 percent for tumors between 2cm and 4cm (153 of 186), and 93.3 percent (28 of 30) for tumors bigger than 4cm, it said.

  “The decreased accuracy of our CNN in an external test set suggested that important differences in CT images might exist between different races and ethnicities, and such diversity needs to be included in the training data for the trained CNN to have good generalisability across diverse populations,” the study said.
• Most Popular

  • 1

    Palmeiras hopes for club cup win

  • 2

    Taiwan to see nine extended holidays in 2026

  • 3

    Taiwan blacklists Huawei and SMIC

  • 4

    US to deliver 10 new F-16s, source says

  • 5

    Ma Ying-jeou’s forum speech regrettable, MAC says