AI diagnoses endometrial cancer with high accuracy

Endometrial cancer is the most common gynaecological cancer in Australia and one of the most diagnosed cancers in Australian women, according to the Cancer Council. Now researchers from Daffodil International University, Charles Darwin University (CDU), the University of Calgary and Australian Catholic University (ACU) have developed an AI model which can detect this cancer with over 99% accuracy, according to a study published in the journal Computer Methods and Programs in Biomedicine Update.

The model, called ECgMLP, examines histopathological images, which are microscopic images of tissue used in disease analysis. The model enhances the quality of the images, identifies the most important areas and analyses the tissue.

“The proposed ECgMLP model outperforms existing methods by achieving 99.26% accuracy, surpassing transfer learning and custom models discussed in the research while being computationally efficient,” said study co-author Dr Asif Karim, from CDU. In comparison, current automated diagnosis for endometrial cancer is reported to be 78.91–80.93%.

“Optimised through ablation studies, self-attention mechanisms and efficient training, ECgMLP generalises well across multiple histopathology datasets, thereby making it a robust and clinically applicable solution for endometrial cancer diagnosis,” Karim added.

Study co-author Associate Professor Niusha Shafiabady, from ACU, said the model also has benefits outside of endometrial cancer diagnosis.

“The same methodology can be applied for fast and accurate early detection and diagnosis of other diseases, which ultimately leads to better patient outcomes,” Shafiabady said.

“We evaluated the model on several histopathology image datasets. It diagnosed colorectal cancer with 98.57% accuracy, breast cancer with 98.20% accuracy, and oral cancer with 97.34% accuracy.

“The core AI model developed through this research can be adopted as the brain of a software system to be used to assist the doctors for decision-making in cancer diagnosis.”

Image caption: The model enhances histopathological images in various ways to highlight the most important area and analyse tissue. Image has been cropped from the original and is courtesy of the study authors under CC BY-NC 4.0