About This Project

Mast cell tumors (MCTs) are the most common malignant skin tumors in dogs (Garrett, 2014.), and their behavior is largely determined by their histopathological grade (cell appearance under a microscope). So far this has been done manually by veterinary pathologists. But, such methods can be time-consuming and subjective. Therefore, it is important to develop an inexpensive and scalable solution that can enable pathologists to grade mast cell tumors faster and more objectively.

Ask the Scientists

What is the context of this research?

Because of the high incidence of MCTs in dogs, their variable biological behavior, and the potential fatal outcome, accurate prognostication of MCTs is critical.

The diagnosis of MCTs is based on skin biopsies. A sample of affected tissue is evaluated and graded by a veterinary pathologist based on the number of mitotic figures, multinucleated cells, bizarre (atypical) nuclei , and presence of karyomegaly (enlarged cell nucleus) (Kiupel et al., 2010).

Deep learning is a fast-growing field that offers novel perspectives for the automation of medical image analysis. It is widely believed that, by automating routine tasks and quantifying various diagnostic parameters, Deep Learning algorithms will lead to one of the biggest transformations in pathology (Colling et. al., 2019).

What is the significance of this project?

By building a deep learning model that will be able to automate histopathological grading, we can make mast cell tumor diagnosis not only faster but also more objective. This will lead to more accurate diagnoses in the future.

Also, with this project, we want to help create a new generation of veterinary pathologists that will have the skills and intrinsic motivation to do the whole end-to-end process needed for creating deep learning models. This is the most sustainable and scalable solution.

What are the goals of the project?

Our main goal is to create a deep learning model for canine cutaneous mast cell tumors histological grading. The model will need to be able to detect and count the number of mitoses, multinucleated cells, bizarre nuclei, and the presence of karyomegaly with >90% accuracy. We also want to develop and open-source a database consisting of at least 10 completely annotated whole slide images (WSI) of canine cutaneous mast cell tumors.

During this process, we will also build two improved versions of our low-cost microscope slide scanner and test not only their hardware capabilities but, more importantly, how the idea of open, programmable hardware can help us build community with the skills and domain expertise needed for creating deep learning applications we need in veterinary histopathology.