Two new algorithms for breast cancer diagnosis

Roche Algorithm, Her2, digital pathology, computational pathology, breast cancer

In January this year, Roche launched two new digital pathology algorithms aimed at improving the diagnosis of breast cancer. These automated algorithms, uPath HER2 (4B5) and uPath Dual ISH, expand Roche’s digital pathology portfolio and provide further tools to assist in determining the most appropriate personalized treatment strategy for individual breast cancer patients.

Why breast cancer?

Globally breast cancer has surpassed lung cancer as the most commonly diagnosed cancer in women, with an estimated 2.3 million new cases in 2020. There are four different sub-types classified according to the presence of hormone receptors and human epidermal growth factor receptor 2 (HER2). Twenty percent of breast cancer cases diagnosed globally each year are found to have the HER2 mutation. Patients with HER2 positive tumor show poor response to chemotherapy and are offered treatment with Herceptin®, a HER2 antibody. This treatment is expensive and as HER2 negative patients do not directly benefit, accurate diagnosis of HER2 expression is critical to avoid risk to patients in prescribing what is an unnecessary, expensive and ineffective therapy. Rapid diagnosis is also important in determining whether the HER2 positive patient will benefit from a targeted therapy strategy.

How are HER2 positive tumors detected?

Usually a biopsy or surgical sample of the cancer is taken from the patient and then tested for tumor biomarkers or genetic make-up using either immunohistochemistry (IHC) or Fluorescent in situ hybridization (FISH) techniques. Histopathology guides the therapy and prognosis of breast cancer and is therefore used primarily in diagnostics. Traditional histopathological slide analysis includes morphological assessment and tumor grading performed by pathologists. However, this process is time-consuming and introduces the risk of inter-observer variability. Diagnostic inter-observer variability has also been reported for HER2 scoring by pathologists. For these reasons, the development of methods to analyse the tissue via digital techniques (computer-aided slide scanning in conjunction with scoring algorithms and artificial intelligence (AI)) has become an important goal. Moreover, image analysis and AI can both improve the diagnostic workflow and detect critical information within the tissue sections which may not be visible to the human eye.

Image analysis in diagnosis

Since pathology is an image-based discipline, accurate analysis, and interpretation of a tissue section image is key to reaching a diagnostic conclusion and computer assisted image analysis has now become an established tool to improve accuracy, reliability, and reproducibility of the results. Image analysis algorithms now routinely support pathologists in making faster, more accurate patient diagnoses in breast cancer.

There are several free image analysis tools available for digital slide images such as ImageJ developed by the NIH. A plugin for ImageJ called ImmunoRatio has been specifically used for the detection of IHC breast cancer biomarkers, ER, PR and Ki67 and a publicly available web application known as ImmunoMembrane has also been used for HER2 IHC. Other commercially available algorithms are available from Tissue Phenomics®, Indica Labs, Visiopharm®, Aperio digital pathology® (Leica), and TissueGnostics® amongst others.

The new Roche algorithms

The uPATH algorithm

uPath HER2 (4B5) is a ready-to-use image analysis algorithm for digital pathology decision support. It helps pathologists to rapidly identify whether tumors are positive for the HER2 biomarker providing consistent and automated analysis. The algorithm suite highlights positively stained tumor cell membranes with a clear visual overlay for easy reference. It helps pathologists in assessing immunohistochemistry and in situ hybridization assays.

uPath HER2 Dual ISH image analysis

The uPath HER2 Dual ISH image analysis algorithm is used to determine HER2 gene amplification or gene status. It is enumerated on the basis of the ratio of the HER2 gene to Chromosome17 in formalin-fixed, paraffin-embedded neoplastic breast tissue specimens. It supports pathologists in defining the areas of interest by providing a heatmap. The algorithm identifies cells to inform the determination of a treatment strategy. Both the algorithms are validated on the VENTANA HER2 (4B5) assay and the VENTANA HER2 Dual ISH DNA Probe Cocktail, and are ready-to-use, integrated within the Roche clinical uPath enterprise software.

Viewpoint

These new Roche algorithms are clearly a useful contribution to the evolving landscape of computer-assisted diagnosis in breast cancer. However, these types of algorithmic tools should be regarded as assisting rather than replacing the decision-making process. It remains up to the Pathologist to make a diagnosis from the information available and generally, image analysis algorithms, although no doubt providing an extra level of security and confidence in the diagnosis, do not replace the pathologist. Algorithms can be relied upon for repetitive work, for screening or counting and for highlighting certain morphological features of interest. The Pathologist provides the experienced, trained eye and the sound judgement.

This is an exciting and rapidly developing field and no doubt the capabilities of these algorithms in conjunction with artificial intelligence and machine learning, will continue to evolve. The ultimate end point may well be unaided computer diagnosis without any human intervention, or perhaps we will instead witness the development of a close synergy between man and machine. The human pathologist can then focus on what he or she does best, providing an experienced perspective based on thousands of previous cases and over a range of staining qualities. The computer partner on the other hand can be relied upon to always give the same judgement under the same conditions. It does not tire or develop human frailties and it does not accidently eliminate regions of the tissue from its observations.

More information on image analysis in breast cancer and the new Roche algorithms can be found in the links below


https://www.roche.com/investors/updates/inv-update-2021-01-11c.htm

https://seer.cancer.gov/statfacts/html/breast-subtypes.html#:~:text=In%202020%2C%20it%20is%20estimated,based%20on%202013%E2%80%932017%20cases.

https://www.uicc.org/what-we-do/thematic-areas-work/breast-cancer#_ftn1

https://pubmed.ncbi.nlm.nih.gov/29175265/

https://healthmanagement.org/c/it/issuearticle/digital-pathology-software-to-improve-breast-cancer-diagnosis-and-therapy-2

Share This Post

Leave a Reply