The multi-institutional team extracted the shape, texture and colour components through artificial intelligence algorithms using different filters.
A team of researchers from Indian Institute of Technology, Guwahati (IITG), along with scientists from renowned research institutes around the world, have designed an automated Artificial Intelligence-based system to detect colorectal cancer using colonoscopy images. Results of the work have recently been published in a prestigious journal belonging to the Nature group – Scientific Reports, according to a release issued by the institute. The team of researchers from IIT, Guwahati was led by Professor Manas Kamal Bhuyan of the Electronics and Electrical Engineering Department, it said. The paper has been co-authored by Dr Kangkana Bora of Cotton University, Guwahati, Dr Kunio Kasugai of Aichi Medical University, Japan, Prof Zhongming Zhao from the University of Texas, Health Science Centre, Houston, USA, and Dr Saurav Mallik of Harvard University, USA have also contributed to the study.
“We have developed an innovative automated system that can help the physician rapidly and accurately detect colorectal cancer from colonoscopy images,” Bhuyan said. This is important because it prevents delays in diagnosis and saves valuable time that can be spent on devising management and treatment strategies for the patient. Assisted by his then-post doctorate student, Kangkana Bora, who is now an assistant professor at the Cotton University, Bhuyan analysed real colonoscopy images generated by Dr Kunio Kasugai of Aichi Medical University, to develop the AI-based cancer detection system, the statement said. During the visual examination, specialists check for the presence and features of abnormal tissue growths (polyps) including shape, surface structure and contour to classify them into different categories (neoplastic and non-neoplastic). The multi-institutional team extracted the shape, texture and colour components through artificial intelligence algorithms using different filters. The statistical significance in the contribution of different components was then evaluated, followed by feature selection, classifier selection based on six measures and cross validation.
“Our extensive experiments show that the proposed method outperforms the existing feature-based (conventional) approaches for colonic polyp detection,” the authors of the paper said. To evaluate the robustness of their system, they compared their work with four classical deep learning models and found theirs to be better than others. The research team is excited with their results and believe that their work would have a global impact in the detection of colorectal cancer. They plan to commercialize the technology in the future as the market need is enormous but before that they have laid out an ambitious research plan to finetune their system. “The work we have reported only focuses on single frames selected by the doctors. In future, we will integrate it with video tracking and automatic frame selection”, Bhuyan said. The team also proposes to implement their analytical approach into a computational tool for easy use, he said.