Using advanced image analytics to spot hidden cancer patterns
Even the experts can have a tough time spotting the early stages of many cancers and other illnesses. That difficulty can cost lives, because the earlier symptoms are detected, the sooner these conditions can be eradicated through drugs, surgery and other treatments.
The advanced medical imaging technologies (MRI, CT, and the like) are not panaceas. That’s because you still need expert humans (physicians and specialists) to visually discern the telltale signs of tumors and other conditions that the images may or may not reveal. And even their trained eyes can miss some non-obvious malignant patterns that are almost indistinguishable from benign tissue.
Spotting non-obvious patterns is what advanced analytics is all about, so it’s no surprise that these technologies (big data, machine learning, statistical modeling and the like) are being applied to medical imaging. This recent article describes advances in imaging analytics that have accelerated the detection of certain forms of cancer. One of the researchers interviewed expressed it well: "Literally, what we're trying to do is squeeze out the information we're not able to see just by looking at an image.”
The success metrics of advanced image analysis are accuracy, speed and scale, compared to what human image analysts can achieve. Though the article doesn’t indicate whether image analytics have passed that test yet, the results so far are encouraging:
- Accuracy: Researchers at Case Western Reserve University achieved 95 percent accuracy in using image analytics on MR images to distinguish between aggressive triple-negative breast cancers, slower-moving cancers and non-cancerous lesions. They also achieved 87.5 percent accuracy in distinguishing among persistent and treatable forms of head and neck cancers caused by exposure to human papillomavirus.
- Speed: The researchers built a mathematical model of how human tissue absorbs a contrast-enhancing dye that can distinguish some malignancies from benign tumors. They then used machine learning and pattern recognition algorithms to diagnose cancers based on telltale texture changes and other evidence that follows from the dye injection. This diagnosis can be accomplished within seconds of the injection, compared to several weeks with biopsies.
- Scalability: The researchers used the tools to analyze MR image data of breast lesions from 65 women. Distinguishing subtypes of breast cancers is always painstaking work, but the researchers used the tools to make quick work of hundreds of gigabytes of image data per patient.
Another benefit from such technology is consistency of diagnosis. A high-performance image-analytics model can allow the medical community to eliminate the inconsistencies among human experts that might otherwise necessitate “second opinions.”