- AI-based contouring tool automates tumour outlining process needed to plan radiotherapy treatment
- Tool helps improve treatment accuracy for disease that predominantly affects the Chinese community
Scientists from the National Cancer Centre Singapore (NCCS) and Sun Yat-sen University Cancer Centre (SYSUCC) in China have collaborated to develop an AI-based contouring tool designed to automate the tumour outlining process that is crucial for planning radiotherapy treatment for nose cancer patients. This AI tool reduces inter-physician biases and inaccuracies that can affect the control of this cancer.
NPC and Tumour Contouring
Nasopharynx carcinoma (NPC), or nose cancer, is the eighth most common cancer in Singaporean men, and predominantly affects the Chinese community. NCCS sees about 150 to 200 new NPC cases a year. NPC is particularly sensitive to radiotherapy, which is therefore the primary treatment for this disease.
Tumour contouring is the most important process in radiotherapy treatment planning. Besides helping radiation oncologists calculate the radiation dose in the designated area, precision in tumour contouring is critical to ensure eradication of the tumour and to avoid toxicities from the treatment. Inaccuracies in tumour contouring and poor quality radiotherapy plans have been shown to affect survival of head and neck cancer patients.
Currently, contouring is done manually by the treating physician. This process can be variable among radiation oncologists as recognition of anatomy and abnormal/normal signs in a tumour can be subjective. The manual process is also time consuming and labour intensive as it entails a thorough review and comparison of the tumour on multiple diagnostic images, such as MRI and CT scans.
In a prospective evaluation of the AI tool, the study found that AI outperforms five out of eight radiation oncologists in terms of accuracy. It also reduced inter-physician variation in tumour contouring by 50%, and cut contouring time by 50% compared to manual contouring. Manual contouring time was 30 minutes compared to 18 minutes with AI assistance.
The AI system imports and draws the entire tumour target and automatically transposes the image to the CT scan. The AI tool is also able to compute a tumour that has spread to other regions.
The findings of the study were published in the journal, Radiology in June 2019. Dr Melvin Chua, a senior consultant radiation oncologist at NCCS and a co-corresponding author of the study, will present the findings at the American Society of Clinical Oncology Breakthrough in Asia Meeting next week. The tool is now being implemented for use in NCCS and several centres internationally as phase 2 of this project aims to investigate the transferability of this AI tool that was developed based on a single large dataset.
Implementation of AI tool in clinic:
Sun Yat-sen University Cancer Centre, which sees a high volume of 600 NPC cases per month, has started using the AI tool in all patients requiring radiotherapy. NCCS is currently involved in a global implementation study to evaluate the impact the AI tool has on the outcome of NPC patients and hopes to implement the AI tool in the clinic if the study shows positive results.