Chapter 87
Cancer — 14
AI has many biomedical applications. Pathology samples are typically stained by hematoxylin and eosin on glass slides and assessed subjectively by pathologists for possible presence of cancer, benign or malignant, and tumor type (organ of origin). That unfortunately leads to some important inconsistency between readers and is rather limited in scope for what now can be achieved with an analysis of digitized whole slide images.
The progress that has been propelled by AI for interpretation of histopathology slides is striking. AI deep neural networks can accurately identify driver genomic mutations, locate the site of origin, articulate the prognosis, predict drug response, and provide treatment recommendations to an extent not previously possible. The accuracy and applications across the full range of cancer types, by both organ and molecular features, has markedly expanded with multimodal AI and the new generative AI (large language models), which now can incorporate data from single-cell sequencing, spatial omics, the tumor microenvironment, imaging, and the immune response. Such a rich depth of information indiscernible to expert human eyes, but “seen” by machine eyes, has transformed how pathology specimens should be assessed. As pathologists Ali Bashashati and S. Larry Goldenberg at the University of British Columbia recently asserted, “AI will not replace pathologists—it will only replace those who do not use AI in their daily practice.”
HEALTH SPAN TREATMENTS
Now that we’ve seen how screening can be upended, how accuracy of diagnosis can be improved, we can see the same principles apply to an overhaul of treatments. For many decades, we’ve heavily relied on toxic chemotherapy that nonspecifically kills our cells that divide frequently, making us sick with gastrointestinal distress, bald, and suffering with chemo-brain. Such treatments are so mutagenic they may even induce a new cancer. We’re on the road to far more intelligent, focused approaches to quash cancer and spare our healthy cells and tissue.
To reiterate, the molecular characterization of a person’s cancer is vital, with so many mutations providing guidance for approved treatments. Remember, there are genes and then there are the proteins they make. Some of these proteins are growth factors that play a critical role in the development of cancer. Genes with important mutations that drive predisposition to cancer include HER2, BRCA1, BRCA2, EGFR, NTRK, ALK, BRAF, ERBB2, FGF, KIT, MET, RET, ROS, PIK3CA, PDGF, PD-L1, and more. Even KRAS, a frequent driver mutation of gastrointestinal and other cancers, once considered undruggable, has FDA-approved treatments that appear to be quite effective. More are on the way, including, perhaps, molecular glues. In patients with advanced non–small cell lung cancer who carry the ALK mutation, five-year survival was improved from 8 percent with traditional therapy to 60 percent using lorlatinib, a pill targeting the ALK and ROS1 genes. Beyond specific mutations, examining the genome sequence for important markers such as microsatellite instability, mismatch repair deficiency, homologous recombination deficiency, and tumor mutational burden provides insight about the use of specific, bespoke FDA-approved drugs. Some of the mutation-targeted drugs have been proven to shrink or eliminate tumors (as observed by scan) but, it must be said, without substantially improving cancer survival.