Background

Lack of cost-effective diagnostics is an important hurdle in the fight against infectious diseases. One of the most widely used tools for diagnosis of the malaria-causative parasite Plasmodium in endemic regions is microscopic examination of blood smears (1). Although microscopy is considered the gold standard; it is labour intensive and time-consuming in practice. Automated robotic microscopes have great potential to enable an era of smart microscopy, but current platforms remain cost prohibitive and pose significant challenges for implementation in resource-poor and field settings such as space, energy, computational power, and deployment demands. In their work, Li et al present Octopi, a low-cost, reconfigurable microscopy platform capable of automated slide scanning (2). They demonstrate the use of the platform by applying it to automated detection of malaria parasites in blood smears. Furthermore, they explore the potential of Octopi for diagnosis of other pathogens relevant for public health in blood, tissue, and sputum, including Schistosomiasis, Leishmania, Trypanosoma brucei, Mycobacterium tuberculosis, Streptococcus pneumoniae and Staphylococcus aureus.

Key findings and developments

Overall development

• Li et al present Octopi, a low-cost, portable, reconfigurable, modular, and automated imaging platform for disease diagnosis in resource constrained settings.
• Octopi can be configured with different disease-specific modules fulfilling different requirements. The modules can be assembled using magnets, which facilitates assembly and reconfiguration.
• Altogether, Octopi is an important advance combining low-cost automated multimodal microscopy and machine learning tools to address the unmet needs for diagnosis of malaria and other diseases.

Specific features of Octopi’smodules

High and low magnification imaging modules

• Octopi has two imaging modules, one with low magnification (low mag module) and one with high magnification (high mag module).
• The low mag module is based on a reverse lens configuration: two multi-element cell phone lenses are used as objective and tube lens in an infinity-corrected configuration. For fluorescence imaging, a small interference long pass filter can be placed between the two lenses. The numerical aperture achieved with the lenses used in this configuration is that of a conventional 10x objective. To add the possibility of auto-focus, motorized focusing was implemented.
• The high mag module uses standard infinity-corrected microscope objectives. For motorized focus adjustment in this module, a combination of a low-cost piezo stack actuator and a standard linear translation stage with extended contact ball bearings/crossed roller bearings was used.

Illumination module

• The brightfield trans-illumination module consists of a LED panel, a diffuser and a condenser. Dark field illumination for low magcan be provided by a ring of LED.
• For fluorescence, oblique angle laser illumination is used, that eliminates the need for a dichroic beam splitter. In commercially available electronics such as pointers, projectors, or Blu-ray/DVD/CD players, direct diode lasers and diode-pumped solid state lasers can provide high optical power, and are available in various wavelengths at low cost.

Scanning module

• A low-cost scanning module was developed, which uses a lead screw linear actuator was used to achieve motorized slide scanning for high throughput imaging.

Control and computation module

• Raspberry Pi, a single board computer with a cost of $35, was chosen as a cost-effective way to control the microscope, and Linux OS was chosen as the operating system to take advantage of open source software packages.
• The platform can support the implementation of low-cost, energy-efficient ASIC chips and optimized hardware for computer vision and machine learning applications.
• Octopi implements image processing and spot detection pipelines on Jetson Nano, which allows real time processing as slides are scanned.

 Automated blood smear examination and detection of malaria parasites

• Given the absence of nuclei in RBCs, fluorescent dyes that bind to nucleic acids can be used to stain for platelets, white blood cells, and parasites. DAPI was used to stain nucleic acids, given its low cost, and temperature stability.
• A two-step processing pipeline was developed for quantification, including background removal and blob detector to identify fluorescent spots of various sizes and intensities. Scanning speed achieved was 1 field of view per second, which allowed visualization of 3 million RBCs per minute.
• At low magnification, segmentation was challenging for cells stained with fluorescent dyes. To overcome the problem, a convolutional neural network was used for object identification.
• Detection of malaria parasites at low magnification is challenging due to the presence of platelets, which appear similar in size and brightness as malaria parasites.
• falciparum parasites have high amounts of RNA, which is a basis for differential detection: fluorescence is red shifted in DAPI-stained RNA compared to DAPI-stained DNA, leading to a differential DNA/RNA ratio. A spectral red shift of 10nm was detected in P. falciparum rings, and was a robust feature for distinguishing platelets from parasites.
• For automatic parasite classification, a boosted tree classifier was built, that uses features from each extracted spot and outputs a class label. The performance of the classifier was assessed by determining false positive and false negative rates.

What I like about this paper

As a parasitologist having worked in field settings, and having done microscopic diagnosis of malaria-infected smears, I think Li et al present a wonderful advance in the form of Octopi with extreme potential for public health applications. As for other pieces of work of the Prakash lab, I find this one of the best examples of using the vanguard of technology, and making it available to everyone, in this case for healthcare. I liked the work because it was very thorough in the design and testing. Moreover, the authors discuss multiple ideas on other implementations that can be done, and how certain features of Octopi were specifically chosen to allow so. This is an example of open science, and open contributions that will allow adaptation of Octopi to satisfy the different needs for pathogen diagnostics.

Open questions

1. In terms of deployment, you discuss you are currently planning a clinical trial for testing the efficacy of the instrument in field conditions. How do you envisage triggering a switch from conventional microscopy to the use of Octopi? And training of personnel?

2. From a parasitology aspect, in your discussion you mentioned that you specifically chose falciparum as the focus of the proof of principle. One of the concerns for diagnosis in all world regions is the identification of strains like P. vivax, and P. knowlesi among others. Are you planning to develop image analysis pipelines that allow further analysis such as correct speciation of different Plasmodium strains? Or detection of gametocytes even in very low numbers?

3. From your setup design, you mention that other imaging modalities that can be implemented include Fourier ptychography, holography/lensless imaging, and LED-matrix and computation-based phase contrast. Can you briefly expand on how these modalities can be used for diagnosis?

4. You briefly mention that besides malaria, you used Octopito identify other parasites like Schistosoma, Leishmania, and Trypanosoma brucei. As well as bacteria like tuberculosis and S. aureus. Some of these pathogens are highly motile and could be identified on this basis. Did you explore Octopi in the context of live imaging or motion detection?

5. In your discussion you mention that besides the field applications, Octopi has also great potential in research settings as it allows parallel units to be used in a single lab to perform super-resolution microscopy, expansion microscopy, spatially resolved profiling of RNA in single cells, and spatial sequencing of single cell transcriptional states in tissues. Have you also tested the performance of modules allowing for each of the techniques mentioned? Do you envisage the creation of more modules to increase the uses of Octopi in the near future?

6. Another form of diagnosis important in field settings is infectious disease vector identification and control (3). Have you used Octopi in this context and do you envisage to do so? For instance, speciation of mosquitoes through different life stages to identify reservoirs? or infection status of different vectors of human and veterinary relevance?

7. By bringing Octopi to all settings in the context of diagnosis, you also provide accessibility to a new form of science to many people everywhere: the different forms of microscopy, image analysis, the concept of machine learning. Was this something you and your lab had in mind when the concept of Octopi first came to you?

References

1. WHO- Malaria and Microscopy – https://www.who.int/malaria/areas/diagnosis/microscopy/en/(Last update 2018)

2. Hongquan Li, Hazel Soto-Montoya, Maxime Voisin, Lucas Fuentes Valenzuela, Manu Prakash, Octopi: Open configurable high-throughput imaging platform for infectious disease diagnosis in the field, bioRxiv, (2019)

3. WHO- Vector control – https://www.who.int/vector-control/en/(Last update 2018)

Tags: diagnosis, epidemiology, microscopy, parasitology

Posted on: 12 August 2019

doi: https://doi.org/10.1242/prelights.13197

(No Ratings Yet)