Deep Learning Improves CT Diagnosis of Gastric Cancer Lymph Node Metastasis

0
0
0

Deep Learning Improves CT Diagnosis of Gastric Cancer Lymph Node Metastasis

Gastric cancer is one of the deadliest cancers worldwide—ranking 5th in incidence and 3rd in cancer-related deaths. In China, it’s a major health threat with high rates of late diagnosis, where cancer cells often spread to nearby lymph nodes by the time patients seek care. Lymph node metastasis is the most common way gastric cancer spreads, and accurate detection is critical: it guides treatment choices like pre-operative chemotherapy and surgical lymph node removal, which directly impact survival.

Traditionally, doctors use enhanced computed tomography (CT) scans to check for metastatic lymph nodes. They look at size (over 10mm), shape (irregular edges, round), and enhancement patterns. But this process is time-consuming, subjective, and prone to errors—especially with large caseloads. Missed or false-positive results can lead to wrong treatment decisions. That’s where artificial intelligence (AI) steps in.

How AI Is Changing Gastric Cancer Diagnosis

A team of researchers from the Affiliated Hospital of Qingdao University and Beihang University Qingdao Research Institute tested a deep learning model called Faster Region-based Convolutional Neural Networks (FR-CNN) to automatically detect perigastric metastatic lymph nodes (PGMLNs)—cancerous nodes around the stomach—on CT scans. Their 2019 study, published in the Chinese Medical Journal, shows how AI can make this process faster and more accurate.

The study was retrospective, analyzing data from 750 gastric cancer patients who had pre-operative CT scans between 2011 and 2018. Of these, 250 had surgery within 2 weeks, allowing doctors to confirm lymph node metastasis with pathology reports (the “gold standard” for diagnosis).

Two Phases of AI Training: From Suspected to Confirmed Cases

The team used a two-step approach to train the FR-CNN:

  1. Initial Training: Three experienced radiologists labeled 1,371 CT images with suspected metastatic lymph nodes using standardized criteria (size, shape, enhancement density). These images—plus 18,780 original CT scans—were fed into the model to learn what cancerous nodes look like.
  2. Precision Training: To fix gaps in the initial model, the team used pathology-confirmed cases. They relabeled 1,004 CT images based on post-operative pathology reports (which definitively say if a node is cancerous) and combined them with 11,340 original scans for further training.

Results: AI Gets Smarter with More Data

The model’s performance improved dramatically after precision training:

  • Mean Average Precision (mAP): Measures how accurately the model identifies target nodes. Rose from 0.5019 (initial) to 0.7801 (precision).
  • Area Under the Receiver Operating Characteristic Curve (AUC): Reflects overall diagnostic accuracy (1.0 = perfect, 0.5 = random). Jumped from 0.8995 to 0.9541—meaning the model correctly identified 95% of metastatic lymph nodes.

This matters because the FR-CNN fixes common human errors:

  • 3D Reconstruction: It views lymph nodes from multiple angles, avoiding missed nodes in 2D CT slices.
  • Continuous Image Analysis: It distinguishes lymph nodes from blood vessels or fat (common false positives) by analyzing sequential scans.
  • Consistency: It doesn’t get tired—unlike radiologists who might miss small nodes after hours of reading scans.

Why This Helps Patients and Doctors

Accurate lymph node diagnosis transforms gastric cancer care:

  1. Better Pre-Operative Chemotherapy: The NCCN guidelines recommend pre-operative (neoadjuvant) chemotherapy for patients with positive lymph nodes. The FR-CNN’s high accuracy helps doctors confirm metastasis quickly, so patients start treatment earlier—improving survival by shrinking tumors before surgery.
  2. Targeted Surgical Planning: Lymph node dissection is the most critical part of gastric cancer surgery. The model shows doctors exactly where metastatic nodes are, so they can remove more targeted nodes (at least 15, per guidelines) and reduce the risk of recurrence.
  3. Reduced Doctor Burden: Reading hundreds of CT scans is exhausting. AI helps radiologists focus on complex cases, cutting down on fatigue-related errors and speeding up diagnosis.

The Bigger Picture: AI and Healthcare Equity

This study also addresses a major gap in Chinese healthcare: uneven access to experienced radiologists. Rural or understaffed hospitals often lack experts to read CT scans accurately. A high-accuracy AI tool like FR-CNN could help these facilities make better diagnoses, reducing health disparities and saving lives.

Conclusion

Deep learning models like FR-CNN are not replacing doctors—they’re augmenting them. By combining radiologist expertise with AI’s speed and consistency, this tool improves the accuracy of gastric cancer lymph node diagnosis, guides better treatment decisions, and eases the load on healthcare providers.

As AI continues to evolve, it could become a standard part of gastric cancer care—catching metastasis earlier and giving patients a better chance at survival.

The original study was conducted by Yuan Gao, Zheng-Dong Zhang, Shuo Li, and colleagues from the Department of General Surgery at the Affiliated Hospital of Qingdao University and Beihang University Qingdao Research Institute.

Trial Registration: Chinese Clinical Trial Registry, No. ChiCTR1800016787; http://www.chictr.org.cn/showproj.aspx?proj=28515.

doi: 10.1097/CM9.0000000000000532

Was this helpful?

0 / 0