Our understanding of disease and the agents that cause it is exploding. More than ever, we can examine a patient’s genes, blood, environment, family history, and personal circumstances to identify and monitor potential problems. But although it’s vital to know as much as possible about a patient’s health, the sheer volume of data poses a challenge in diagnosing disease and selecting the best possible treatments. Add to this the growing shortage of pathologists and it’s clear that a new approach is needed.
This is particularly true in prostate cancer – a field that has often faced controversy over the “best” way to detect and characterize the disease. Does the biopsy show cancer? If so, what type and at what stage? Which is the best option for treatment? Answering these questions can be difficult, time-consuming, and controversial – but, with the advent of artificial intelligence (AI) diagnostics, there is a light at the end of the tunnel. The AI advantage Deep Bio, a company dedicated to improving diagnostic processes and expanding access to quality.
The AI advantage
Deep Bio, a company dedicated to improving diagnostic processes and expanding access to quality healthcare, sees AI as the way forward. Their goal? To make medical diagnoses more reliable and reproducible, while reducing turnaround times and easing the burden on pathologists and laboratories worldwide. That’s why they’ve chosen to make their first target prostate cancer – a disease of many challenges.
The company’s first solution is DeepDx-Prostate, a system that uses a trained deep learning model to spot acinar adenocarcinoma of the prostate on scanned hematoxylin and eosin slides. The tool’s basic function is to identify slides that show evidence of cancer, allowing pathologists to concentrate their attention on these cases – a streamlining effort that allows more cases to be processed in less time, reduces the overall workload of pathologists, and alleviates the mental and physical fatigue of long hours spent hunched over a standard microscope.
But the AI doesn’t stop there; it can offer even more detailed information. How? By using Gleason scoring, the system can identify areas of interest, highlight tissue regions with different colors, and provide results including:
- Gleason score
- relative proportions of each Gleason pattern
- percentage involvement of cancer
- representative lesion images taken from the slide
It can even generate pre-filled pathology reports containing the score and representative images to save even more time and effort.
Deep Bio’s DeepDx-Prostate tool in action, highlighting different tissue areas for Gleason scoring; pattern 3 in yellow, 4 in orange.
Knowing the score
Gleason scoring can be tricky. DeepDx-Prostate can help by providing not only automatic scoring, but also a breakdown of how that scoring was achieved, complete with images. Different colors highlight areas with features representative of different Gleason scores, allowing pathologists to take a birds’-eye view of a case and make quick decisions. Company founder and CEO Sunwoo Kim said, “Pathology reporting guidelines have recommended to particularly include the percentage of Gleason pattern 4, as it provides clinically significant information on predicting patients’ prognosis (1).” Currently, the model is expected to achieve a similar level of agreement to human pathologists – and, with more data, it’s designed to
become even more accurate.
For consensus annotation of complicated cases, the DeepDx-Prostate model can be deployed with all of the functionalities of the local tools – but on the Internet, so that pathologists can share, annotate, and discuss slides with colleagues. The system is in preparation for an open beta trial next year as DeepDx Connect - Prostate. Together, these tools support pathologists in the laboratory and help them move toward Deep Bio’s ultimate goal – to make high-quality pathology portable and accessible to all.
DeepDx Connect - Prostate is registered as in vitro diagnostic in compliance with the CE. MFDS Korea approval is in progress. US market entry is expected to begin in 2020.
For Research Use Only. Not for use in diagnostic procedures.
For more information, contact [email protected]
- Biospectator, “Deep Bio launches AI-diagnostic software on prostate cancer ..‘grants certifications’” (2018). Available at: bit.ly/2pDSelP. Accessed October 3, 2019.
- H Chang et al., “Automated Gleason scoring of prostate needle biopsy images using deep neural networks and its comparison with diagnoses of pathologists”. Poster presented at USCAP 2019; March 16–21, 2019; National Harbor, USA. Abstract #811.
