AI and lung cancer prognosis
To follow up on an earlier post on the future of artificial intelligence, AI has been making serious inroads in radiological imaging for a while. Unsurprisingly, histological imaging is the next frontier, and AI is conquering that as well.
A subset of lung cancer patients will see metastatic spread to their brains. A recent study reports that deep learning algorithms can distinguish which cancers will, and which won’t, metastasize to the brain based on histology of primary tumor sections. Furthermore, the computer predictions are more accurate than trained pathologists. From the abstract:
“The DL algorithm results were compared to a blinded review by four expert pathologists. The DL-based algorithm was able to distinguish the eventual development of brain metastases with an accuracy of 87% (p < 0.0001) compared with an average of 58.3% by the four pathologists and appears to be particularly useful in predicting brain metastases in stage I patients. The DL algorithm appears to focus on a complex set of histologic features. DL-based algoithms using routine H&E-stained slides may identify patients who are likely to develop brain metastases from those who will remain disease free over extended (>5 year) follow-up and may thus be spared systemic therapy.”
The study started as an attempt to discover predictive biomarkers and that failed. Instead, AI was able to make predictions about cancer progression using biopsy samples that were already being collected for diagnosis.
AI-guided histopathology predicts brain metastasis in lung cancer patients
The Journal of Pathology – March 4
This is a great post Joel. Have you ever seen this: http://www.galleri.com?
@Lj,
Thanks! Not familiar with galleri per se, but the number of cancer biomarkers is growing. A former grad student is working for a company that discovers and commercializes cancer biomarkers.
A search for biomarkers for lung cancer metastatic potential was what started the study I posted about. AI/DL instead found predictive patterns in H&E staining.
Does AI Help or Hurt Human Radiologists’ Performance?
Harvard Med School – March 19
It Depends on the Doctor. New research shows radiologists and AI don’t always work well together.
Heterogeneity and predictors of the effects of AI assistance on radiologists
Nature Medicine – March 19
@Dobbs,
From the discussion section:
“Our findings, based on a large-scale sample of 140 radiologists, highlight the existence of radiologist heterogeneity in treatment effects, which has substantial implications for both absolute and relative performance.”
This is like discovering water at the bottom of the ocean. What would be remarkable is if, among 140 human radiologists, there was zero radiologist heterogeneity.
Also too:
“Collaboration between clinicians and AI developers, focusing on personalized strategies and continuous improvement of AI models, will be essential for achieving the full potential of clinician–AI collaboration in healthcare.”
I could have told them this before the study began.
The paper is useful in documenting the nature of the discordance between different radiologists and AI models. But already AI is improving radiology and it will only get better.