Detecting polyps and lesions during colonoscopies

• Colorectal cancer is the second most common cause of death from cancer for men and women alike.
  
• In most cases, colonic polyps are symptomatic of the early stages of colorectal cancer.
  
• Regular screening helps detect polyps at an early stage and thus prevent this type of cancer from developing.

Problem

Such polyps, however, are often missed during colonoscopies. Small polyps and flat new polyp formations that are deeply embedded are especially difficult to detect. And the physician’s experience and level of attention can also influence the detection rate.

Solution

Automatically detect and highlight polyps in the endoscope’s live video signal.  

Goal

Increase polyp detection sensitivity and specificity using artificial intelligence.

Current status

• Built Multi-center, multi endoscope database
  
  •     “Bayreuth” DB: 2484 640 x 504 HD (no open access database!)
    
  •     Public dataset CVC-ClinicDB: 612 388 x 284 SD
  •     Public dataset ETIS-LaribPolypDB: 194 1225 x 966 HD

• Trained Deep Neural Network to detect and outline polyps
• Results for above datasets: recall of 83-92% and precision of 74-86%
• Looking for commercialization partners

----------------------------------------------------------------------------------------------------------------------------------------------------

Literature

Zobel et al, Computer Aided Detection of Polyps in Whitelight-Colonoscopy Images using Deep Neural Networks , Current Directions in Biomedical Engineering 2018;4(1)