Detecting colon cancer in general abdominal CT scans, i.e. without bowel emptying performed before colonoscopy, is extremely challenging. With nowadays technology, it is not possible to automatically detect cancer in the large intestine without prior preparation of the intestine as required in colonoscopy.

A study has revealed that more than 20% of malignant findings in the colon are missed on general abdominal CT scans, indicating a need for tools to help the radiologist in this challenging task. 
The researchers – Prof. Nahum Kiryati from Tel Aviv University and Dr. Arnaldo Mayer from Sheba Hospital, have successfully developed AI-based algorithms that enable automatic segmentation of the large intestine in general abdominal CT scans, that is, scans done without emptying, enema, or any preliminary “preparation” of the intestine. This step is particularly challenging due to the very large variation in the appearance of the colon that has not undergone “preparation”, and is a necessary step towards the automatic identification of suspicious findings in the colon. The researchers developed a system based on neural networks for the automatic identification of suspicious findings for malignancy in the large intestine in general abdominal CT scans, without any preparation of the intestine. The system and technology was validated using clinical abdominal CT scans in cooperation with the imaging institute at the Sheba Hospital: general abdominal CT scans were sent directly from the scanners to a server in the hospital’s computing cloud, processed automatically, and the processing results, including automatic marking of pathologies detected in the large intestine, were transferred to the archive of the hospital, ready for the radiologist’s review.

UNMET NEED
Millions of abdominal CT scans are performed each year around the world for many reasons unrelated to colon cancer. Detecting colon cancer without bowel emptying performed before colonoscopy, is extremely challenging. Since CT scans already exist, Incidental detection of colon cancer in these scans can save many lives. Colon cancer lesions are missed in over 20% of the abdominal CT scan.
       
OUR SOLUTION
The researchers developed AI-based algorithms that enable automatic segmentation of the large intestine in general abdominal CT scans. This step is particularly challenging due to the very large variation in the appearance of the colon that has not undergone “preparation”, and is a necessary step towards the automatic identification of suspicious findings in the colon. The researchers developed a computerized system based on neural networks for the automatic identification of suspicious findings for malignancy in the large intestine in general abdominal CT scans, without any preparation of the intestine. The system and technology was validated using clinical abdominal CT scans in cooperation with the imaging institute at the Sheba Hospital: general abdominal CT scans were sent directly from the scanners to a server in the hospital’s computing cloud, processed automatically, and the processing results, including automatic marking of pathologies detected in the large intestine, were transferred to the archive of the hospital, ready for the radiologist’s review.

Sample results of the automatic colon segmentation

APPLICATIONS
Automatic incidental detection of colon lesions in abdominal CT.

STATUS
Fool proof of concept laboratory scale, with promising experimental results.

A joint research of Dr. Eliahu Konen and Dr. Arnaldo Mayer from Sheba Medical Center.

INTELLECTUAL PROPERTY
NEURAL INCIDENTAL DETECTION OF COLON LESIONS IN ROUTINE ABDOMINAL CT, PCT/IL2022/050518, Priority Date – 18 May 2021