Advertisement

Google is working on AI for ultrasound diagnosis and cancer therapy

Its medical AI language model also scored 85 percent on a doctor-level medical exam.

zz/John Nacion/STAR MAX/IPx

AI isn’t just good for writing term papers or clickbait financial explainers; it could help save lives in the medical field. At Google’s annual The Check Up healthcare event, it announced AI-related partnerships for ultrasound readings, medical language models and cancer treatments — areas where the technology could someday serve as a force for good.

Google sees AI as crucial in reading ultrasound devices in regions without enough trained specialists. Although the sensors are more accessible than ever, they require experts to conduct exams and interpret images. The company’s AI models could help simplify that process by identifying data like the early detection of breast cancer and gestational age in expectant mothers. To help make that a reality, the search giant is partnering with Kenya nonprofit Jacaranda Health to research AI-based ultrasound treatments for mothers and babies in government hospitals. “Through this partnership, we’ll conduct exploratory research to understand the current approach to ultrasound delivery in Kenya and explore how new AI tools can support point-of-care ultrasound for pregnant women,” said Google’s Health AI head Greg Corrado and Engineering VP Yossi Matias in a blog post today.

The company is also working with Chang Gung Memorial Hospital in Taiwan to research how AI can detect breast cancer via ultrasound as an alternative to mammograms, which have limited availability in lower-resource regions. Mammograms can also be less effective in populations with higher breast density.

Dr. Shelly Tien, 40, performs an ultrasound on MC, 24, who is in her second trimester, at Planned Parenthood in Jacksonville, Florida, March 15, 2022.   REUTERS/Evelyn Hockstein.         SEARCH
Evelyn Hockstein / reuters

In other areas, Google says its medical-focused large language model (LLM) has improved significantly. Med-PaLM 2, the company’s next-generation healthcare LLM, recently scored 85 percent on doctor-level medical exam questions — an 18 percent improvement from the previous version’s score. “This model not only answered multiple choice and open-ended questions accurately, but also provided rationale and evaluated its own responses,” said Corrado and Matias.

However, don’t expect a ChatGPT-like bot to replace your doctor anytime soon, as Google cautions the technology still isn’t ready for real-world work settings. For example, an evaluation on criteria like scientific factuality, precision, medical consensus, reasoning, bias and harm found “significant gaps” when answering medical questions. Corrado and Matias noted, “We look forward to working with researchers and the global medical community to close these gaps and understand how this technology can help improve health delivery.”

Google has also partnered with Mayo Clinic to explore AI’s part in planning radiotherapy for cancer treatment. The research focuses on reducing the tedious and time-consuming steps of the radiotherapy process — most notably, “contouring.” This technique requires clinicians to draw lines on CT scans to separate cancerous areas from nearby healthy tissues that the radiation could damage, a process that can take up to seven hours for one patient. The company says it will soon publish research from the three-year study while formalizing an agreement with Mayo Clinic to explore more radiotherapy-based research, AI models and commercial uses.

Finally, Google sees AI as helping with chest x-ray screening for tuberculosis. The company is partnering with an AI-based organization to make AI-powered TB screenings widely available in Sub-Saharan Africa. Its partners have committed to donating 100,000 free screenings to help detect tuberculosis early and provide early treatment to reduce its spread.