Wednesday 05 March 2025
Deep learning has revolutionized many fields, from image recognition to natural language processing. But one area where it’s particularly powerful is in medicine – specifically, in the analysis of digital pathology slides.
Pathologists have long relied on manual examination of tissue samples under a microscope to diagnose diseases like cancer. But this process can be time-consuming and prone to human error. By applying deep learning algorithms to these images, researchers hope to make diagnoses faster, more accurate, and less dependent on human interpretation.
A new paper presents a toolkit called SpinPath that aims to simplify the reuse of deep learning models in digital pathology. The authors developed SpinPath as a response to the lack of standardized frameworks for sharing and reusing these models. They argue that by making it easier to access and apply pre-trained models, they can accelerate research and improve patient care.
SpinPath is designed to be user-friendly, with a Python-based interface that allows researchers to quickly evaluate different models on their own data. The toolkit includes a collection of pre-trained models, known as the SpinPath Model Zoo, which are hosted on Hugging Face Hub – a popular platform for sharing machine learning models.
To test SpinPath, the authors applied nine foundation models (pre-trained deep learning models designed for general-purpose feature extraction) to an external dataset called SLN-Breast. They found that each model had its own strengths and weaknesses, with some performing better than others in terms of accuracy and speed.
The fastest model, Phikon, took just 40 seconds per slide to complete, while the UNI model achieved the highest balanced accuracy (0.975) – a measure of how well a model performs across different types of slides. The authors also developed a JavaScript-based browser tool that allows users to perform specimen-level inference with no coding required.
SpinPath’s potential benefits are twofold. First, it could streamline research in digital pathology by making it easier to access and evaluate pre-trained models. This could lead to faster development of new diagnostic tools and more accurate diagnoses for patients. Second, SpinPath could help democratize access to deep learning technologies, making them more accessible to researchers without extensive machine learning expertise.
The authors acknowledge that there are still challenges to overcome before SpinPath can be widely adopted. For example, they note that some models may require significant computational resources, and that the toolkit is currently limited to analysis of whole-slide images.
Cite this article: “SpinPath: A Toolkit for Simplifying Deep Learning Model Reuse in Digital Pathology”, The Science Archive, 2025.
Digital Pathology, Deep Learning, Medical Diagnosis, Image Analysis, Spinpath, Pre-Trained Models, Machine Learning, Hugging Face Hub, Foundation Models, Whole-Slide Images
