The aim of the PICCOLO project has been the development of a fully functional wide-field fluorescence based, OCT and MPT photonics endoscope for improving colorectal cancer diagnosis providing in-vivo image-guided biopsy capabilities and higher sensitivity and specificity than current diagnostic methods.

PICCOLO consortium has worked through all the project pursuing these objectives. On the hardware side, enhanced wide-field imaging and probe integrating OCT and MPT technologies have been delivered. On the software side, a computer aided diagnosis (CAD) system that makes use of advanced deep learning models for automatic image diagnosis of wide-field and OCT/MPT images has been delivered. Additionally, a method that virtually stains MPT images into gold-standard Haematoxylin-Eosin (H&E) images with high sensitivity and specificity to easy image interpretation has been developed. Prototype and algorithms have been fully tested on murine models and validation on human models has been (unfortunately) partially achieved.

Individual results are described into detail in the results catalogue here. (Note:  updated metrics are available for some the CAD algorithm as progress have been made after catalogue release).

The delivered probe fits the working channel of a colonoscope and it’s no exclusive of a specific device. The colonoscopy industry can benefit from this design solution and offer the optical biopsy probe as an additional component that can be sold separately from the colonoscope, hence maintaining current base retail price. Additional value can be offered if CAD system algorithms are added to the probe purchase. Health systems, clinicians and patients can also benefit from this, as it would make possible to update existing hospital colonoscopes with least monetary investment. Besides, optical biopsy adoption would lead health systems to reduce diagnosis costs, time and patient trauma.

Advanced image processing methods based on deep learning have been implemented on the CAD. State of the art deep learning techniques have demonstrated to reach precision rates comparable to experts, surpassing them in some cases. Until recently, CAD clinical software was based on classical machine learning techniques that in most cases were not reliable enough to be adopted on the daily clinical routine. With the introduction of deep learning, the market forecast for clinical software is expected to rapidly growth in the short term. In this sense, the resulting metrics of the different algorithms developed in the project, suggests that the realization of the optical biopsy is closer than ever. Besides this, during the PICCOLO project, it has been demonstrated that the future of clinical devices relies on the combination of hardware and software. The virtual staining algorithm approach proposed at the project, which automatically transforms MPT images into gold-standard H&E images, can facilitate the adoption of new imaging techniques by clinicians and at the same time easy the dreaded learning curve.