Meeting the challenges of diagnostic surgical pathology

1. Tissue acquisition

Existing methods

Most biopsies are performed by hand with or without imaging equipment, depending on the location and visibility of the tissue and area of concern. For example, the use of imaging equipment would be unnecessary in the case of skin cancer, as the lesion is usually visible. For lesions that cannot be seen or felt through the skin, a surgeon can perform an image-guided biopsy utilising imaging methods such as CT, MRI or ultrasound, to locate the lesion and collect a tissue sample using a special needle that pierces through the skin. The alternative, more invasive method is to perform a surgical incision to locate the suspicious lesion and remove it, which is usually done for larger samples.

It should be noted that there is a trade-off between invasiveness and diagnostic accuracy. Smaller biopsies, such as needle biopsies, tend to be less invasive but carry greater risk of missing a lesion which leads to a false-negative result – known as a sampling error. Meanwhile, larger biopsies, such as an excision biopsy, are less prone to sampling errors, yet can carry higher complication risks and a longer recovery time for the patient. Regardless of biopsy type, all biopsy samples are then sent to a pathology lab for full examination.

One of the challenges with handheld imaging probes and biopsy instruments is that they are positionally uncalibrated, which can lead to human error in determining the position of the lesion relative to the instrument. For example, during a freehand prostate biopsy, the urologist manually handles the TansRectal UltraSound (TRUS) probe to determine the position of the prostate and uses a needle to collect a sample.

Emerging technology – the role of surgical robotics in biopsy

Emerging robotic technologies have the potential to offer less invasive biopsies, without sacrificing accuracy. Considering the example above, a robot-controlled TRUS probe can perform precise scanning motions, producing uniformly distributed 2D images, suitable for 3D reconstruction. This allows for automatic and accurate alignment of the needle guide to the target tissue. A study revealed that the robot’s mean targeting error was nine times lower than a urologist’s. Robotic prostate biopsy was also associated with a higher significant cancer detection rate of 43%, compared to 36% achieved by freehand prostate biopsy.

There is currently no surgical robot on the market which relies solely on real-time TRUS imaging. The Biobot iSR’obot Mona Lisa performs prostate biopsies using a pre-operative MRI scan in conjunction with real-time ultrasound imaging. However, this approach can be time consuming, costly and is less suitable for local anaesthetic due to longer intervention time. While the performance of a TRUS-guided prostate biopsy robot has only been explored within academic research to date, it has huge potential for commercialisation.

Specifically designed equipment can access hard to reach areas within the body without making an incision. One example is robotic bronchoscopy, a minimally invasive procedure that allows surgeons to reach lesions within the lungs through the airway, utilising a robotic-guided flexible bronchoscope. By using a robot, surgeons can avoid the risk of human fatigue, while enabling them to precisely control the arm with great repeatability.

While there are a variety of different robotic bronchoscopy platforms, each with their own end effectors (the device at the end of the arm) and navigation systems, most still rely on a CT scan for initial mapping of the lungs. For example, The Monarch™ RB platform uses an articulating bronchoscope within an articulating sheath, each controlled by separate robotic arms, along with an integrated camera and an electromagnetic navigation system. The Ion™ RB system uses a single ultrathin bronchoscope equipped with shape-sensing technology to localise and maintain scope position. When using these platforms, serious complications are rare and recovery is fast, with patients usually being able to leave the same day as the procedure.

The challenges of surgical robotics

Despite the benefits they bring, the upfront cost of buying a surgical robot can be astronomical and may not be a realistic option for hospitals lacking in funding or patient demand. Hiring or training surgeons to be qualified enough to operate these highly specialised robots can also be challenging and extremely time consuming. Maintenance and sterilisation may also require additional infrastructure, adding onto cost and complexity.

Some robot platforms also currently lack clinical and practical evidence of how they provide benefits over existing procedures, or justification for the increased costs they introduce. For example, with robotic bronchoscopy, although localisation and complication rates have improved, the results have been mixed for diagnostic yield.

CT to body divergence is also a technical issue for many robotic platforms. This arises from the difference between a static pre-procedure scan and natural organ/muscle movement during the procedure. This issue remains unresolved by many surgical robot platforms which use a pre-procedural CT scan as a “map” for the robot to reference for navigation, which introduces errors in navigation path and end location. More work is needed to allow these surgical platforms to adapt and respond to live changes during a procedure. Feasible options to achieve this include introducing new real-time imaging methods, using multi-modal sensing or AI driven imaging.

Developing surgical instruments that are compatible with imaging techniques for real-time imaging can also prove challenging, especially for real-time MRI guided biopsies, as ferromagnetic components cannot be used. As such, many conventional materials, actuators and sensors often used in robotic modules are no longer an option. However, a study has revealed that accurate sampling rates of non-robotic MRI guided biopsies have shown to be superior to US guided biopsies for prostate biopsies. If the benefits of robotic handling can be integrated with MRI real-time imaging, this could yield significant benefits in terms of sampling accuracy and diagnostic yield. An emerging field which has the potential to address this challenge is soft robotics, which utilises unconventional materials and soft actuators to function.

2. Specimen preparation

Existing methods

Specimen preparation consists of several steps, some of which require dexterous manual handling by a pathologist. The first step is “grossing”, which involves an examination of the tissue sample by the naked eye, followed by manipulation of the specimen to select the most diagnostically valuable parts, to determine priority for processing and report generation. The specimen is then placed into a labelled cassette for identification.

For many pathology labs, the cassettes are placed within a processor inside a chamber, where different chemical solutions are displaced in and out of the chamber for fixed periods of time. This includes formalin solution for fixation, alcohol for dehydration, xylene for clearing and wax for infiltration. In general, the whole process takes a couple of hours and is usually set up to run overnight. The result is a more sturdy and robust tissue sample, that is suitable for handling and maintaining structure during subsequent steps.

The following step is embedding, where the processed tissue is moved from the cassette and embedded into a metal mould filled with molten paraffin using an embedding station, such that the paraffin block created after cooling encases the processed tissue. Specific to each biopsy type, there may be an orientation and placement the sample must be in within the mould to result in an accurate examination. The biopsy sample is typically handled carefully using tweezers, to avoid generating artifacts.

After this comes sectioning, considered to be the most destructive step. This involves using a microtome to slice the tissue embedded paraffin block into extremely thin sections and mounting them onto glass slides. The most widely used microtome is the rotary microtome, where the displacement of the paraffin block towards or away from the cutting knife is manually controlled using a hand wheel. The plane of slice is parallel to the bottom surface of the metal mould, highlighting the importance of tissue orientation in producing the most informative cross-sections. The slides are stained to enhance contrast and highlight certain cellular components for diagnosis.

I’ve had the pleasure of shadowing pathologists at work in the embedding lab, where they often work non-stop embedding as many biopsy samples as possible, whilst maintaining a high level of standard and care. Everything matters during this process, the pressure they apply on the tissue sample to keep it as flat as possible without damaging the sample; making note of any under-processed tissue for re-processing and identifying underlying structures to determine appropriate orientation (for example skin biopsy orientation such that the epidermis is cut last during sectioning). These are only a handful of all the considerations pathologists make per tissue sample they embed. This manual process is often labour intensive and again can be prone to human error.

3. Medical diagnosis

Existing methods

In traditional methods for medical diagnosis using surgical pathology, pathologists examine stained slides under a light microscope to identify or diagnose diseases. Each tissue sample produces multiple slides which are all essential for diagnosis. The pathologist may choose to annotate the slides before viewing them and upon first examination, will view the slide at low magnification to examine the overall architecture. The pathologist will then switch to high magnification to further scrutinise suspicious areas. For example, at low magnification the pathologist may be able to identify that the patient has cancer, and at a higher magnification they are able to identify the stage of the cancer.

Some features of the sample may require extensive searching, as the microorganism or cell in question may be so small that it is invisible at low magnification, however higher magnification means there is more ground to cover – a bit like “Where’s Wally”, but if Wally was hidden somewhere within dozens of pages, rather than a single spread. In less straightforward cases, the pathologist may need a second opinion. Whilst this could be as simple as grabbing another colleague within the lab for a quick discussion, it could be more involved such as shipping the slides to a different lab for another team’s opinion.

Once the slides are thoroughly examined, the pathologist has identified enough features and has gathered sufficient evidence to prove their hypothesis, the process of producing a report begins. The report includes patient information, details about the biopsy and its location, biopsy information obtained at the grossing stage, a microscopic description and the final diagnosis detailing the disease and its extent. This information is useful in determining the urgency of the patient case and planning their treatment.

Emerging technologies for surgical pathology diagnosis – digital pathology

One of the key emerging technologies for this step of the diagnostic surgical pathology process is digital pathology. Digital pathology is an established yet not widely applied field within surgical pathology, which aims to transfer all the information gathered during the pathology process onto a digital platform, in order to streamline the acquisition, management, sharing and interpretation of the biopsy sample.

Through whole-slide scanning (WSI), microscopic details on glass slides are captured and transformed into high-resolution digital images, facilitating remote viewing and analysis without the need for light microscopes. For example, the Philips IntelliSite Pathology Solution WSI scanner has a resolution of 0.25 μm per pixel and a batch scan capacity of 300 slides. The scanner also allows for image processing, annotating and sharing using proprietary digital pathology software, enabling collaboration among pathologists worldwide.

One emerging and attractive benefit of digital pathology lies in its capability to provide a platform to collect the data essential for training machine learning (ML) algorithms, as well as to implement these trained ML algorithms to aid in diagnosis. By leveraging the wealth of data collected by pathology labs, encompassing scanned slides, tissue types, patient demographics and resulting diagnosis, ML models can be trained to identify disease patterns and provide valuable insights to pathologists. While there are numerous regulatory challenges of AI in healthcare, there are examples of AI being used in pathology. Paige Prostate Detect is the first FDA approved pathology AI, which focuses on prostate cancer. Developed using over 55,000 slide scans from over 1,000 institutions around the world, it boasts a 70% reduction in cancer detection error and 65.5% reduction in diagnosis time.

The integration of AI-powered decision support has immense potential for accelerating diagnosis and improving accuracy to cope with the growing demand for pathology services.

The challenges of digital pathology

Despite the benefits of digital pathology, the initial investment cost of implementing it may not be feasible for many pathology labs. These costs include software related expenses, specialised lab equipment, powerful hardware and supporting infrastructure. WSI scanners themselves are composed of several specialised, high-fidelity components such as a trinocular microscope with robotic control of illumination intensity and a high-resolution camera, all of which are expensive pieces of equipment.

If software processing and data storage is done in-house, high-capacity and fast-processing hardware will also be required to process and store scanned slides, as each image file is enormous, typically 1-3 GB per image. High-capability hardware is essential for AI/ML software to operate reasonably fast, especially to avoid slow processing times which would defeat the purpose of introducing a new tool intended to expedite the procedure. Using a cloud network resolves the need for such hardware, whereby powerful cloud-based servers can carry out the computing required externally, though this comes at the price of direct access over data and version control. Since all the data is stored remotely, if cloud outage occurs, access to all data could be blocked. Cloud computing also raises important concerns around patient data privacy and safety, as third party organisations are responsible for hosting these cloud networks and could be susceptible to cyber attacks or data leaks.

Successful implementation of WSI and AI/ML greatly relies on the quality of scanned slides, while high quality slides rely on consistent and well controlled specimen preparation. Thick sectioning can result in slides failing to scan, while artifacts in the specimen, the presence of bubbles or under and over staining, can lead to poor AI-generated advice affecting the final diagnosis.

Developing robust, accurate and unbiased ML models also heavily relies on access to large amounts of standardised, high-quality, demographically diverse data. This data may be limited by privacy regulations and data protection policies, such as GDPR compliance and a lack of adoption of existing standards such as DICOM. This makes it especially difficult to develop AI for rare forms of cancers or diseases, where many examples are needed for accurate machine learning.