Skip to main content

Deep learning AI in cancer diagnostics

Periodic Reporting for period 1 - AiforCancerDX (Deep learning AI in cancer diagnostics)

Reporting period: 2019-11-01 to 2020-10-31

Pain point – Aging populations and increasing cancer rates are constantly increasing the number of samples in healthcare. Despite the growing trend of digitalisation, pathologists are still reviewing the samples manually, which is a subjective, time-consuming, expensive manual process that exposes patients to the risk of misdiagnosis. Thus, there is a clear need for an affordable tool that supports and streamlines the workflows of pathologists, complementing humans with automated and objective analysis.
Solution – Aiforia Technologies Oy is revolutionising pathology. Our Aiforia® Cloud platform augments the efficiency and consistency of pathologists’ clinical sample analysis. It removes the slow, manual, and inconsistent workflow by automatically performing a range of laborious image analysis tasks, in a fraction of time with unprecedented accuracy and consistency. This is enabled by our disruptive Deep Learning AI software, implemented on a cloud platform, and built specifically for pathology applications.
To achieve the overall objective of the project - to scale up the operations of Aiforia® Cloud and become the global leader in automated tissue sample analysis – several actions have been performed.
Deliverable 1.1 Deep Learning AI algorithm for PD-L1 in lung cancer
Training of the first version of the algorithm is completed and it is ready for testing. The progress of the deliverable has been described in more detail in Report D1.1.

Deliverable 1.2 Deep Learning AI algorithms for prostate cancer diagnostics
Training of the first version of the algorithm is completed and it is ready for testing. The progress of the deliverable has been described in more detail in Report D1.2.

Deliverable 1.3 Deep Learning AI algorithms for tumour length and volume measurements
Training of the first version of the algorithm is completed and it is ready for testing.

Deliverable 1.4 Platform optimisation for clinical use cases
Aiforia has made substantial progress in optimizing its platform for clinical use cases.

Deliverables 2.1 - 2.4. Validation of the algorithms in a clinical laboratories and regulatory approvals
Aiforia has completed extensive preparatory work to ensure the implementation of the validation plan and CE-IVD documentation, and application of the CE-IVD mark within the 2nd period.

Deliverable 3.1 Business Development Report
Aiforia has made substantial progress with respect to business development activities. The company has strengthened its salesforce and increased its international presence and lead generation activities as planned. The company will submit the Business Development Report as defined in the project plan in month 24.

Deliverable 4.1 Project Management Report
Aiforia project management activities have contributed to the overall progress of the project. The company will submit the Project management report in accordance with the project plan in month 24.
The development of the Deep Learning algorithm for PD-L1 in lung cancer (D1.1) was started from scratch by obtaining image data required for training of the algorithm. The algorithm was created by training convolutional neural networks (CNN) by labelling a set of training images with the Aiforia® Create software tools. The objective is to overcome the challenges of the current PD-L1 scoring solutions and to develop and certify the world’s first ready-made Deep Learning AI algorithms to make cancer diagnostics efficient and accurate. This algorithm is ready to enter the next phase (WP2: testing, validation and regulatory approvals).

The development of the prostate cancer diagnostics algorithm (D1.2) was completed. The algorithm it is ready for the subsequent testing and clinical validation phase. The algorithm was created in collaboration with the Helsinki University Hospital by labelling a set of training images with the software platform. The initial results were promising and the target sensitivity and specificity ratios of 95% were met.

The third deliverable (D1.3) consists of algorithms for automating prostate cancer tumor length and volume measurements. These tools are technically ready and their functioning in the case-specific context will be verified when the prostate cancer algorithms are finalized.
Lastly, the fourth deliverable (D1.4) the optimized Aiforia platform is showing strong progress. As the first step, Aiforia developed a viewer functionality enabling pathologists to review digitized samples from computer screens. Second, an application programming interface (API) has been developed enabling the exportation of result data. This empowers the clinicians and researchers to use the AI-generated data in combination with other datasets and contributes to the integrations laboratory information management systems. Finally, Aiforia is currently developing the viewer functionality to make available a CE-marked device for viewing the AI-generated overlay results in the viewer.

The second work package (WP2) has also been started. The preparations for validating and obtaining the regulatory approvals are progressing as planned with the partner hospitals. Practical arrangements have been made with the Helsinki University Hospital to validate the models developed in the project. After the validation step, obtaining the required regulatory approvals will enable Aiforia to offer its customers a solution that surpasses the level of the current state of the art.

All in all, the development and commercialization measures have supported Aiforia in maintaining its position as one of the global leaders in digital pathology. The industry is developing at a fast pace and players in the digital pathology field are aiming to offer integrated AI analysis systems for clinical clients. Ultimately, when ready and certified in line with our plans, Aiforia’s algorithms will reduce the pathologists’ workloads, and contribute to more accurate and faster turnaround of patient samples. The patients will benefit from this by receiving better diagnoses faster, and in a way that optimizes their treatments and improves the treatment outcomes.