Posted on May 01, 2024

Dr Cristian Scatena is a consultant pathologist at the University Hospital of Pisa, Italy, and assistant professor of pathology at the University of Pisa. His specialty is breast cancer and dermatopathology, particularly melanoma.

We spoke to Dr Scatena about recent work he presented at the European Breast Cancer Conference in Milan, using MammaTyper® to predict the response to neoadjuvant chemotherapy in HER2-positive breast cancers. We also discussed the role that artificial intelligence will have in pathology in the future.**

Thanks for speaking to us Cristian. Firstly, what challenges are you facing as a pathologist when it comes to neoadjuvant chemotherapy in breast cancer?

Our labs and our breast units constantly see more and more cases of breast cancer that need neoadjuvant chemotherapy, particularly for triple-negative and HER2-positive breast cancer.

But among these patients we see that approximately 40 to 50% do not achieve a pathological complete response. We know that the pathological complete response is a surrogate of the outcome for a patient.

Since I’m a pathologist, my interest is to give the oncologists a tool that may help them to identify before starting the neoadjuvant chemo if that patient will achieve a complete pathological response or not, with the aim of adjusting the treatment.

You presented some of your research with MammaTyper® at the recent European Breast Cancer Conference, can you summarise that work and what you found?

We have been working on the expression of HER2 protein by immunohistochemistry (IHC), and so we are familiar with the pattern of expression of the HER2 protein. With this study, we wanted to see if MammaTyper could predict the response to neoadjuvant chemo.

We enrolled 49 patients with HER2-enriched tumours and we tested the preoperative biopsy with MammaTyper. Some were oestrogen-negative, but others were triple-positive, so positive for the expression of ER and PgR.

Using a machine-learning Python-based decision tree algorithm, we could stratify and clearly identify those patients that had a complete pathological response from the others [without pCR].

In terms of sensitivity, specificity, positive predictive value, and negative predictive value, this Python-based decision tree algorithm is quite strong.

What was the most surprising aspect of the results?

I did not expect that even with this small number of cases the sensitivity and specificity would be so high. If you have this statistical accuracy with such a small number of patients, this is very, very good, because it will probably be better with a bigger number of patients.

At first, I was convinced that the mRNA expression of HER2 [on its own] could give us an answer. But we saw that this is not the case. As we saw, Ki67 and progesterone receptor are very important among the four markers in the MammaTyper test.

What are your plans for the future with this work?

In order to give it more power, we are now enrolling new patients to make sure the two groups – one with complete pathological response and the one without – are equal in numbers and so they are comparable.

We know this algorithm works based upon MammaTyper results, but it would be interesting to understand if IHC could be used for the same purpose – at present we do not know.

I hope that we will have a strong tool that can help our breast units, our multidisciplinary teams, to give the right patient the right therapy.

What difference could this make?

It’s a part within the larger concept of personalised medicine. This work could mean that MammaTyper could discriminate patients with HER2 positive breast cancer that will achieve a pathological complete response from those who will not. This may represent for our patients in the near future a powerful decision tool in terms of de-escalation or escalation treatment approaches – in the neoadjuvant setting, but maybe also in the adjuvant setting.

What role do you see artificial intelligence playing in pathology in the future?

The machine-learning and the artificial intelligence tools are able to see things that we do not see. In this case and in other cases where we apply artificial intelligence to pathology, they are of course more sensitive and more specific. When you have to find very small differences in an IHC, the human eye is not the good thing to use.

I would stress the importance of getting familiar with artificial intelligence. Pathologists are not so familiar yet with artificial intelligence, because we are used to things that we see with our eyes.

But this is also a problem of culture, of course, and a problem of different generations. Because I see in my residents that they have a completely different approach with everything digital. So I think that for the pathologists of the future, we will see an implementation of artificial intelligence and digital pathology in the pathology labs.

Also, there are always less and less pathologists. The pathology units are overwhelmed with specimens, with tests to do, and very few pathologists. Artificial intelligence and digital pathology will be very useful to have the right diagnosis at the right time.

Are you interested in conducting studies into predicting response to neoadjuvant treatment? Get in touch to discuss a collaboration with Cerca Biotech: