If clinicians detect ovarian cancer before it metastasizes, the survival rate is greater than 90%. The problem is that clinicians diagnose ovarian cancer using serum biomarker detection combined with transvaginal ultrasonography, but these methods cannot detect the cancer early enough to reduce mortality.
To design a more sensitive ovarian cancer diagnostic tool, Daniel Heller, a biomedical engineer at Memorial Sloan Kettering Cancer Center and author of a new study, and his team developed a nanosensor based on single-walled carbon nanotubes (1). When they paired this nanotube sensor with a machine learning model, they detected ovarian cancer with greater sensitivity than current methods and identified new biomarkers for the disease, potentially allowing for earlier cancer detection and better health outcomes.
“Hysterectomy, where they remove the ovaries, fallopian tubes, and the uterus, is a good preventive surgery, but it also causes premature menopause. If you can actually just detect early, that would be the best. Then those kinds of procedures wouldn't have to be done nearly as often,” said Heller.
Heller’s carbon nanotube-based sensor, or “molecular nose,” uses a perception-based strategy to detect more molecules than the number of sensors it has, akin to how humans differentiate many more smells than we have smell receptors. The sensing of a combination of different molecules by smell receptors leads to a unique smell sensation.
Carbon nanotubes look like tiny stretches of chicken wire rolled into a tube. To detect different ovarian cancer biomarkers, Heller’s team modified their nanotubes with small chemical groups called organic color centers (OCC) and then wrapped them with single-stranded DNA molecules. The DNA wrapping facilitates suspending the nanotubes in a liquid solution such as blood serum and enhances the nanotubes’ molecular selectivity. When molecules in the local environment interact with OCCs, the distinct infrared wavelengths emitted by the OCCS can be measured with a spectroscopic apparatus.
Heller’s team collected information from their carbon nanotube sensors and trained a machine learning model to distinguish between healthy and ovarian cancer samples.
While the combination of a sensitive and specific nanosensor without the machine learning model enabled differentiation between high-grade ovarian cancer samples and healthy ones, the system could not differentiate high-grade ovarian cancer from other disease conditions, but it could do so when machine learning was applied, differentiating conditions such as endometriosis, low-grade ovarian cancer, and non-gynecologic cancers. The sensor technology performed better than the current clinical screen, a combination of longitudinal cancer biomarker 125 (CA-125) and second-line transvaginal ultrasonography.
“I suspect that if they had a combination of this nanosensor with CA-125 and transvaginal ultrasonography, they might discover that their detection rate here is close to 100%, which is better than the current CA-125 biomarker assay and transvaginal ultrasonography,” said Ray Iles, the chief scientific officer at the clinical diagnostics company MAP Sciences who was not involved in the study. “Don't look at it in isolation. Look at [it] in combination.”
Although Heller and his team envisioned their new sensor solely as a detection method, it may also have biomarker discovery applications. When the researchers individually eliminated each spectroscopic variable from the machine learning analysis to understand how it affected the model’s performance while accounting for any interference from drugs or other chronic conditions, they discovered that some serum markers that were not previously considered ovarian cancer biomarkers were more important to the model performance than the known biomarkers.
Heller and his team hope to develop this technology for detecting early-stage ovarian cancer and to determine if this technology can be used to detect other cancer types, especially those that can be cured when detected early.
“This kind of technology moves in a direction parallel to self-driving cars and other technologies that use AI and machine learning. One could continue to train this technology to detect more and more things and to get better and better,” said Heller.
- Kim, M. et al. Detection of ovarian cancer via the spectral fingerprinting of quantum-defect-modified carbon nanotubes in serum by machine learning. Nat Biomed Eng 6, 267-275 (2022).