How Franklin.ai created training data batches 95% faster with V7 Darwin

How Franklin.ai created training data batches 95% faster with V7 Darwin

 Agar plate used to diagnose infection
Franklin AI logo

Franklin.ai helps pathologists diagnose with confidence and report faster.

Industry

Healthcare

Use Case

Digital pathology

Company size

11-50

Headquarters

Haymarket, Australia

Data labelling
Discover our solutions

About

Franklin.ai‘s goal is to address a global shortage of pathology specialists by providing pathologists with a digital assistant that uses state-of-the-art technology to enhance diagnostic accuracy and efficiency.

This Australian-based joint venture between Harrison.ai and Sonic Healthcare will provide pathologists with AI solutions that act as a second pair of eyes. These solutions have the potential to improve diagnostic accuracy and increase workflow efficiency, quality of care, and outcomes.

Time needed to create batches of 10,000 images

from 10 hours to 30 mins

"The configurability and extensive API interface of V7 has enabled us to quickly create, customise, deploy and make changes to our labelling processes all through automated communications with our own backend tooling."

Benjamin Johnston

Principle Engineer at Franklin.ai

File type

WSIs

Favorite feature

Integrates with your S3 infrastructure

Time needed to create batches of 10,000 images

from 10 hours to 30 mins

"The configurability and extensive API interface of V7 has enabled us to quickly create, customise, deploy and make changes to our labelling processes all through automated communications with our own backend tooling."

Benjamin Johnston

Principle Engineer at Franklin.ai

File type

WSIs

Favorite feature

Integrates with your S3 infrastructure

Time needed to create batches of 10,000 images

from 10 hours to 30 mins

"The configurability and extensive API interface of V7 has enabled us to quickly create, customise, deploy and make changes to our labelling processes all through automated communications with our own backend tooling."

Benjamin Johnston

Principle Engineer at Franklin.ai

File type

WSIs

Favorite feature

Integrates with your S3 infrastructure

The challenge

Franklin.ai’s mission is to meet a worldwide shortage of pathologists by building AI digital pathology models capable of performing diagnostic tasks with increased accuracy and efficiency.

To build highly accurate AI models, Franklin.ai needs to create training data from thousands of specialist Whole Slide Image (WSI) files, annotating anomalies within tissue cells on a pixel-perfect level.

However, without the right tools managing, storing, and annotating these enormous data files would divert significant engineering time and resources.

The solution

Franklin.ai uses V7 Darwin to train segmentation models that identify anomalies in tissue cell WSIs to a near-perfect degree of accuracy, helping pathologists make confident diagnoses worldwide.

Collaboration between V7 and Franklin.ai to develop custom tiling code significantly decreased the storage costs involved in the data annotation process and facilitated more efficient collaborations between teams.

Furthermore, V7’s direct integration with Franklin.ai’s private AWS S3 storage bucket has enabled them to annotate data with a global team of users without patient data ever leaving their private storage bucket.

The results

Using V7, Franklin.ai has deployed its first product to market, helping pathologists save time by using AI to analyze medical data.

By streamlining data management and annotation with V7, this team decreased the time needed to create batches of 10,000 images to their training data pipeline from 10 hours to 30 mins.

Training data needs

Creating pixel-perfect annotations for WSI files involves sharing enormous files between expert medical labelers.

Manually managing this sharing would divert significant engineering resources to downloading, uploading, and storing files resulting in poor-quality and costly training data.

To solve this, this team needed a platform that would integrate with their existing AWS storage, and enable them to keep all data within their own S3 bucket.

So, this team needed a training data platform capable of supporting high-quality annotations, hosting large amounts of data, and facilitating easy collaboration between annotation experts.

Testimonial

Testimonial

Testimonial

"Without the V7 platform I struggle to think how we could complete our labelling projects. We have a very large labelling team that is spread throughout the globe. The configurability and extensive API interface of V7 has enabled us to quickly create, customise, deploy and make changes to our labelling processes all through automated communications with our own backend tooling. This is all possible while also supporting digital pathology files that are often gigabytes in size. The ability to store and serve all of our labelling data through our own S3 infrastructure has also meant that we maintain the required control of our data at all times as well as the costs of hosting such large datasets. This is a very unique, but also essential feature set that V7 provides."

"Without the V7 platform I struggle to think how we could complete our labelling projects. We have a very large labelling team that is spread throughout the globe. The configurability and extensive API interface of V7 has enabled us to quickly create, customise, deploy and make changes to our labelling processes all through automated communications with our own backend tooling. This is all possible while also supporting digital pathology files that are often gigabytes in size. The ability to store and serve all of our labelling data through our own S3 infrastructure has also meant that we maintain the required control of our data at all times as well as the costs of hosting such large datasets. This is a very unique, but also essential feature set that V7 provides."

Benjamin Johnston

Principle Engineer at Franklin.ai

Benjamin Johnston

Principle Engineer at Franklin.ai

Benjamin Johnston

Principle Engineer at Franklin.ai

Building intelligent AI to assist pathologists requires training machine learning models to identify abnormalities in cell tissues from WSIs.

Franklin.ai chose V7 Darwin to create this training data after careful consideration.

Key factors in their decision were our stringent data security policies designed to help build compliant medical AI products and V7’s support in building customer tiling code to overcome the challenges of large file sizes common with medical specialist WSI data.

Franklin.ai was attracted by the ability to integrate V7 Darwin with their existing AWS S3 storage, saving time-sharing files. Our accessible UI also spared the diversion of engineering resources needed to develop a custom front end.