|More women in India die from cervical cancer each year / Photo Credit: designer491 (via Shutterstock)|
More Indian women die from cervical cancer than in any country, reported Manish Singh of TechCrunch, an American tech news platform. Every year, cervical cancer kills around 67,000 Indian women, which is more than 25% of the 260,000 deaths worldwide. Early detection and effective screening are helpful in reducing cervical cancer, but the testing process of detecting it is time-consuming.
Unfortunately, the existing methodology that cytopathologists employ is in itself time-consuming. At SRL Diagnostics, India’s largest chain to offer pathological and radiological diagnostic services are using AI. Last year, Microsoft collaborated with SRL Diagnostics to create an AI Network for Pathology to ease the load of cytopathologists and histopathologists. The chain receives over 100,000 PAP smears samples annually. About 98% of the samples are normal, while 2% of them require intervention.
Cytopathologists at SRL Diagnostics analyzed digitally scanned versions of Whole Slide Imaging (WSI) slides manually, marking their observations afterward. Each one comprises of 300 to 400 cells. These slides were used as training data for Cervical Cancer Image Detection API. Dr. Arnab Roy, Technical Lead for New Initiatives & Knowledge Management at SRL Diagnostics noted that there is also subjectivity involved in examing the slides, which is possibly linked with the expert’s experience.
Principal Applied Researcher at Microsoft Azure Global Engineering Manish Gupta said that AI could “identify areas that everybody was looking at" and “create a consensus on the areas assessed.” Thousands of tile images of cervical smear were annotated by cytopathologists from multiple labs and locations. SRL Diagnostics and Microsoft’s work bore fruit as the AI model can now distinguish between normal and abnormal smear slides accurately, according to Microsoft’s official news site. The slides are under validation in labs for three to six months. The AI can also classify and base smear slides on the seven-subtypes of the cervical cytopathological scale.
Dr. Roy said that the API has the potential to boost the productivity levels of a cytopathology section by about four times. He added, “In a future scenario of automated slide preparation with assistance from AI, cytopathologists can do a job in two hours what would earlier take about eight hours!”