CHICAGO (GenomeWeb) – Machine learning in genomic medicine and molecular diagnosis got a boost last month when Sophia Genetics announced that its technology now can scan medical images to predict the evolution of solid tumors. By combining this analysis with genomic data to classify tumors, Sophia believes it can help clinicians make better diagnoses and treatment decisions.
The Swiss genomic analytics company said at the recent Precision Medicine World Conference in Silicon Valley that the Sophia AI artificial intelligence platform can analyze quantitative features from a consecutive series of PET scans, CT scans, MRIs, x-rays, and other "standard-of-care" medical images to help determine how a tumor might change in terms of size, volume, and location. Sophia said that it had already proven the concept with about 1,000 cases of lung cancer, kidney cancer, glioma, and meningioma.
"Anytime a PET scan, a CT scan, or an MRI is being taken by a radiologist, what the radiologist needs to do is see where [the solid tumor] is located and have an idea of how the solid tumor will evolve," said CEO and cofounder Jurgi Camblong. Particularly with glioma and meningioma, "It's very important to have an idea of how quickly these tumors will grow because of course they will compress the brain," Camblong said. Physicians base decisions of whether or not to operate on expected growth.
"If, on subsequent CT scans and MRIs, it appears that the tumor itself doesn't match the prediction, this is a strong sign that the tumor is evolving, adding additional mutations to its molecular pattern, which should be a motivation for the pathologist [to order] a new molecular test to see what are these new mutations and eventually adapt the treatment before it is too late," Camblong continued.
"The mathematical models that are behind it are inspired from the mechanics of fluids."
This month, Sophia also said that it had received a CE-IVD mark for a molecular diagnostics test designed to improve detection of leukemia and other hematological diseases. The test is a sequencing-based assay combined with data analytics to detect mutations that cause myeloid and lymphoid leukemia.
But the big news is around the imaging. Sophia claims to be the first commercial vendor to marry genomics and radiomics with artificial intelligence.
"Thanks to this algorithm and machine learning on imaging, we can predict the evolution of the tumor," said spokesman Tarik Dlala. "This, combined with genomic information, is helping clinicians make faster decisions."
Molecular data can, for example, lead to a determination if a glioma is oligodendroglioma, astrocytoma, or glioblastoma, Dlala added.
IBM's Watson has been applying machine learning to imaging ever since Big Blue bought medical imaging informatics company Merge Healthcare for $1 billion in 2015, but it is not to that level yet. "Watson for Genomics does not incorporate any image analytics, and I'm not aware of anything we've done that combines genomic data with imaging data," an IBM spokeswoman told GenomeWeb.
The Sophia technology is the result of 12 years of work, predating the founding of the company in 2011, according to Chandra-Mouli.
"The big motivation was the oncology workflow of patients for basically diagnosis that starts with a PET scan, a CT scan, or an MRI. It's very important to follow longitudinally the patient in terms of diagnosis and of treatment," Camblong said.
When it started rolling out its AI platform in 2014, Sophia decided to focus on genomics, Camblong said, but "with the ambition of adding additional layers of data." The radiomics capability represents one of those layers.
"We thought [now] was the right time for adding an additional layer because of our ability of scaling up, and because of the maturing of [clinical genomics] practice in the hospital," Camblong said. In the last two years, hospitals have started to "understand the benefit of data-driven medicine in oncology."
About 35 percent of the 10,000 cases clinicians use Sophia to diagnose each month are in oncology, according to Camblong. Other applications are for congenital conditions including autism, metabolic disorders and various heart ailments, he said.
Camblong said that this extra layer of information, combining radiology with molecular data, leads to more accurate diagnoses and smarter treatment decisions. The Sophia system scans images in search of patterns.
Camblong said the AI has been "validated" in several hundred cases. It is "capable of predicting the growth of a solid tumor according to two subsequent images that have been taken over time," he explained. "It helps the surgeon, the pathologist, the radiologist, the chemotherapist, to adapt the treatment path."
The new technology —limited to solid tumors for now — is in use at five European hospitals. Camblong named four of them: University Hospital of Bordeaux, France, Humanitas Research Hospital in Milan, Italy, Hôpital Tenon in Paris, and Bergonié Institute in Bordeaux.
"The idea is to roll out this technology as a product and start with our AI right from the first diagnosis in these four applications" (lung cancer, kidney cancer, glioma, and meningioma) to improve predictions of how cancers might grow, Camblong said. "And then combine this imaging data with the molecule information. We believe that over time this will give us a better ability of further predicting how such types of tumors in other patients are going to evolve."
As with any AI, the technology gets better the more information it has. "You need basically imaging data and the segmentation of the tumor [after running it through the Sophia platform] and the expert eye of the radiologist. With that, you can train the system to become better and better," Camblong said.