Patho-ML, a low-cost, rapid, point-of-care, breast cancer diagnostic tool

 

Authors: Kyle Lee, Kyle Mani, Megan Maniar, Joseph Nguyen

 

Contact: kl781@scarletmail.rutgers.edu

 

Importance: More than half of cancer deaths occur disproportionately in low- and middle-income countries where a lack of pathologists causes delayed treatment and poor patient outcomes. While early-stage breast cancer is treatable, effective screening programs and efficient diagnostic services are essential. In Botswana, a major barrier is the number of pathologists available to read tissue slides. This results in increased turnaround time between biopsy collection and diagnosis, which in turn reduces treatment options and long-term prognosis.

 

Objective: To develop a rapid, point-of-care, diagnostic tool to reduce turnaround time between biopsy collection and diagnosis to improve low-income patient outcomes.

 

Evidence Review: An effective solution to these issues must be cost-effective and easy to implement. Costly imaging tools like mammography, lengthy training programs, and major structural changes are not feasible. We have developed Patho-ML, a rapid, point-of-care diagnostic tool, for breast cancer. After a breast biopsy is conducted and conventional H&E-stained tissue slides are prepared, our tool images these slides using a cell phone camera attached to a standard microscope. The images are then input to a machine learning algorithm for automated tissue classification. Additionally, the infrastructure needed for the tool’s implementation, including cell phones, microscopes, and networking capabilities, is readily accessible at the major free and charitable clinics, which serve low-income patients.

 

Findings: Our current machine learning algorithm based on the VGG16 neural network has been successfully trained on over 10,000 histopathological images. With this prototype, we have been able to classify breast tissue images with 97% accuracy, 98% sensitivity, and 96% specificity.

 

Conclusions and Relevance: Patho-ML targets developing nations, which serve patients unable to attain care, and where access to pathologists is limited. Our approach can be applied to free clinics across the country, and can be used on different pathological diseases. Ultimately, our approach will decrease turnaround time to just a few hours for these vulnerable populations.