Customer Success Story: Vistapath | Scale AI

Overview

Vistapath is Building the Next-Generation Pathology Lab

A 42-year-old patient suspected of having prostate cancer had a biopsy taken. The patient's doctor sent the biopsy sample to a pathologist for a diagnosis. The diagnosis was positive for an aggressive form of cancer, leading to the removal of the patient’s prostate. As is standard procedure, the removed prostate was sent back to the lab for further analysis, where pathologists discovered that the prostate had no cancer cells. It turns out the histology lab mistakenly swapped this patient’s biopsy sample with another patient's during a process called grossing. Grossing is the process of assessing and documenting the physical characteristics - size, color, consistency, and features of that tissue - in a report. Pathology technicians may also elect to orient, ink, or dissect that tissue as part of the grossing process. A pathologist then uses the gross report for diagnosis. By leveraging computer vision and artificial intelligence, Vistapath increases quality, accuracy, and efficiency across critical lab processes like grossing, building the next-generation pathology lab to reduce these types of mistakes. Vistapath began by asking the question: is grossing best done by a computer, or a person? They quickly realized that the answer is a bit of both. Some steps, such as moving and slicing tiny tissue samples are done better by humans, as we are much more dexterous than robots. However, when it comes to examining tissue samples and quickly extracting key characteristics such as dimension, color, and spotting and measuring lesions to populate a gross report, computer vision systems can complete this step more efficiently and accurately.

The Problem

Vistapath Needed Annotation Tooling Their Histologists Could Easily Use

In order to improve the safety and efficiency of grossing, Vistapath developed the Sentinel, a device that sits on the grossing workstation and augments histologists’ work with a tissue detection model. The labs Vistapath serves can use a variety of histology cassettes and background substrates.

The technical team at Vistapath is concerned with building and training computer vision models that are robust to all types of variations in appearance. In order to do this successfully, the data scientists working on these models need hundreds to thousands of images accurately annotated and processed efficiently in order to improve the quality of their algorithms.

Given the expertise needed to annotate images of tissue samples, Vistapath knew they needed an annotation tool their histologists and experts could easily use. Vistapath began by using an open-source annotation tool to label the tissue samples. The open-source tool, however, lacked automation and could not effectively scale with Vistapath’s needs.

The Vistapath team then tried a tool with more automation, such as human-in-the-loop segmentation algorithms to accelerate annotation. This other vendor, however, could not meet Vistapath’s security and compliance requirements. The team at Vistapath needed a partner who not only provided the annotation automation tools they needed to accelerate labeling but also could meet their strict security and compliance requirements.

“Using Studio has also allowed us to iterate on our taxonomy very easily. The ability to quickly label new data on new taxonomies has been very smooth and has led to model improvements.”

Daniel Sturniolo

Sr. ML Engineer

Vistapath

The Solution

Scale Studio Gives Vistapath the Automation and Auditing Tools They Needed

Vistapath chose to partner with Scale and use Studio because "Scale is the known name-brand. Nobody is going to question us from a security or compliance perspective when working with Scale," explained Jacob Guggenheim, VP of Engineering at Vistapath. In addition to meeting Vistapath's compliance requirements, Studio also provided the automation tools to help their internal labelers and specialists label data more efficiently and auditing tools to help reviewers confirm the accuracy of annotations. "The team that does the labeling like Studio a lot, and we are confident that the data we send to Scale is in safe hands," added Guggenheim.

The Result

Vistapath is Improving Pathology Labs and Enhancing Patient Experience

Since partnering with Scale, the Vistapath team saw an accelerated rate of annotation and increased productivity for their ML team. “Using Studio has also allowed us to iterate on our taxonomy very easily. The ability to quickly label new data on new taxonomies has been very smooth and has led to model improvements,” said Daniel Sturniolo, Sr. ML Engineer at Vistapath. Beyond serving as Vistapath’s annotation tooling, “Scale Studio also serves as a data warehouse for our team, providing a central location where datasets are stored and can easily be queried,” added Guggenheim.

As Vistapath's tissue detection model continues to improve with more and better data, histologists and pathologists can expect faster and more accurate grossing reports to improve diagnostics and enhance patient outcomes.

“Scale Studio also serves as a data warehouse for our team, providing a central location where datasets are stored and can easily be queried.”