Can artificial intelligence determine circumferential resection margin status in patients with rectal cancer?

One of the most important variables evaluated prior to therapeutic assignment in patients with rectal cancer is the status of the circumferential resection margin. The correlation between circumferential resection margin positivity and local recurrence has been repeatedly proven. One of the 20 standards of the new American College of Surgeons Commission on Cancer National Accreditation Program for Rectal Cancer is performance of a high quality rectal cancer protocol MRI interpreted by a synoptic report reviewed prior to therapeutic decision making with the entire multidisciplinary team. One of the challenges in interpretation of any image is variability among interpreters. Computer algorithms involving deep learning, neural networks, and artificial intelligence have played a role in many aspects of medicine. Wang and colleagues from Qingdao, China evaluated 240 patients with rectal cancer during a two year period. The authors utilized a training group of 192 patients 66.1% of whom were male with a mean tumor distance of 6 cm from the anal verge and a mean tumor size of 5.5 cm. They employed a validation group of 48 patients 66.7% of whom were male in whom tumors of a median size of 5.2 cm diameter were situated from a median of 6.2 cm cephalad to the anal verge. The authors relied upon standardized axial, coronal, and sagittal sections in diffusion weighted images as well as in fast spin echo T2-weighted images. Using a regional network detailed in their figure 2, they were able to achieve accuracy, sensitivity, and specificity of circumferential resection margin status of 0.932, 0.838, and 0.56, respectively. These images were automatically recognized in 0.2 seconds. The study design and the results indeed intriguing.  Imaging appears to this non-imager to be well suited to neural network analysis. This type of algorithm might be particularly helpful in interpretation of rectal cancer protocol high quality MRI’s in institutions in which the rectal cancer imagers are not in abundance. If this platform or one like it is proven to be reproducible around the world, it will allow patients to avail themselves of the best possible interpretation regardless of the setting in which they are being treated. I congratulate the authors upon this outstanding and very intriguing concept.