According to the American Cancer Society, the diagnosis of pancreatic cancer is poor, with a survival rate of only 12%. Computed tomography (CT) emerges as a cornerstone for effective detection methods, recognized for its widespread use and sensitivity in identifying pancreatic cancer. However, despite its wider adoption, CT scans miss nearly 40% of tumors <2 centimeters, underscoring the pressing need for a revolutionary tool to augment pancreatic cancer detection.
In a recent article titled ‘Pancreatic Cancer Detection on CT Scans with Deep Learning: A Nationwide Population-based Study,’ published in Radiology, author Po-Ting Chen and his team demonstrated that the deep learning (DL)-based tool enabled accurate detection of pancreatic cancer on CT scan with reasonable sensitivity. The comparative analysis involved subjects diagnosed with pancreatic cancer and those with a normal pancreatic profile, weaving a comprehensive tapestry of data to refine the determining capabilities of the AI tool.
The study aimed to create and authenticate a tool utilizing (DL for identifying pancreatic cancer in CT scans. Researchers retrospectively gathered contrast-enhanced CT studies from patients diagnosed with pancreatic cancer between January 2006 and July 2018. These were compared with CT studies of individuals with a healthy pancreas (control group) acquired between January 2004 and December 2019.
Chen and colleagues analyzed 733 control participants randomly assigned to training, validation, and test sets, and 546 pancreatic cancer patients (mean age, 65 years ± 12 [SD], 297 men). In this study, the DL tool achieved 95.9% specificity (141 of 147; 95% CI: 91.3, 98.5) in the internal test set and 89.9% sensitivity (98 of 109; 95% CI: 82.7, 94.9) in comparison with the original radiologist report (96.1% [98 of 102]; 95% CI: 90.3, 98.9). There was no significant (P =.11) difference in sensitivity between the two sets of data.
Using a test set of 1473 actual CT investigations from Taiwanese institutes (669 cancerous, 804 control), the DL tool distinguished between CT malignant and control studies with 89.7% sensitivity (600 of 669; 95% CI: 87.1, 91.9) and 92.8% specificity (746 of 804; 95% CI: 90.8, 94.5) (AUC, 0.95; 95% CI: 0.94, 0.96), with 74.7% sensitivity (68 of 91; 95% CI: 64.5, 83.3) for malignancies <2 cm.
In line with this finding, a retrospective study by Liu et al. found that pancreatic cancer can be reliably identified on portal venous CT images using DL techniques like convolutional neural network (CNN). As a computer-aided diagnostic tool, the CNN model has the potential to help radiologists and physicians identify pancreatic cancer.
The integration of artificial intelligence into medical imaging not only augments the capabilities of radiologists but also opens avenues for more effective and timely diagnosis, especially in the field of oncology. The synergies between technology and healthcare may contribute to more precise, accessible, and impactful diagnostic practices.
References
- Chen PT, Wu T, Wang P, Chang D, Liu KL, Wu MS, et al. Pancreatic Cancer Detection on CT scans with Deep Learning: A Nationwide Population-based Study. Radiology. 2023 Jan;306(1):172–82.
- Liu KL, Wu T, Chen PT, Tsai YM, Roth H, Wu MS, et al. Deep learning to distinguish pancreatic cancer tissue from non-cancerous pancreatic tissue: a retrospective study with cross-racial external validation. The Lancet Digital Health. 2020 Jun 1;2(6): e303–13.