Real-time AI assistance led to much greater efficiency and lower cost, in a study by Dutch researchers of sentinel node biopsies for breast cancer. Sentinel lymph nodes (SNs) are the first lymph nodes to drain lymphatic flow from the tumor. Testing them can be a difficult and time-consuming process.
This team found AI-assisted pathologists were much less likely to use the tedious process of immunohistochemistry (IHC). They also saw substantive cost savings (~3,000 €), significant time reductions, and up to 30% improved sensitivity for AI-assisted pathologists.
Their study appeared in Nature Cancer June 27, and was led by Carmen van Dooijeweert, MD, department of pathology, University Medical Center Utrecht.
“At this point it is unthinkable that AI would diagnose cancer without any oversight from a trained pathologist, not in the least because of the ethical and legal consequences if its diagnosis would be incorrect,” van Dooijeweert told Inside Precision Medicine.
She added, “So, we believe augmented intelligence, or AI-assistance, is the only way for AI-use in healthcare, there needs to be a human in the loop, which patients can trust.”
This is an ideal task for AI. SN slides have to be assessed diligently, but a key question is how many need to undergo the tedious process of IHC analysis. Macrometastases are pretty easily spotted, but smaller metastases are not, and may require IHC. However, only about a third of SNs contain metastases, so a lot of these more expensive tests are not necessary. But the results have big implications. Patients with an SN containing metastases, without prior neoadjuvant treatment, usually require adjuvant treatment, whereas those with a negative SN or only ITCs do not.
In this single-center trial, the team prospectively investigated the relative risk of IHC use per detected case of SN metastasis using an AI-assisted clinical workflow for the detection of metastases in breast cancer. Their main objective was to determine whether the AI-assisted workflow could reduce the material and personnel resources spent on IHC stains, while maintaining work safety standards.
A total of 190 SN specimens from 182 participants were included, of which 100 were included in the AI arm (52.6%) and 90 were included in the control arm (47.4%). Over 40% of participants received neoadjuvant therapy, nearly always consisting of at least chemotherapy. They found lower relative risk of IHC use for AI-assisted pathologists, with subsequent cost savings of ~3,000 €.
The team writes, “Currently, AI algorithms have already been developed for various tasks such as tumor detection, tumor subtyping, (tumor) biomarker evaluation and tumor grading… Yet, prospective studies on actual clinical implementation are lacking, sometimes even many years after promising publications.”
The SN procedure is performed in breast cancer patients in whom diagnostic imaging is negative for involved axillary lymph nodes and entails a combination of intratumor injections with radiocolloid and a perioperative injection of patent blue to detect and resect the SN(s).
The presence of metastases in the SNs is strongly associated with worse survival and consequently guides treatment according to the size of the metastases (that is, macrometastases (≥2 mm), micrometastases (<2 mm) or isolated tumor cells (ITCs; single tumor cells or tumor cell clusters with a maximum diameter of ≤0.2 mm and a maximum number of 200 cells per section). In general, patients with an SN containing (micro)metastases, without prior neoadjuvant treatment, require adjuvant treatment, whereas those with a negative SN or only ITCs do not.
For pathologists, the assessment of these SNs is a tedious and labor-intensive task with a dichotomous answer: the presence or absence of SN metastases. Subsequently, the SN slides have to be assessed diligently so as to not miss small but clinically relevant metastases. Meanwhile, the overall yield is low because approximately two-thirds of SNs do not contain metastases.