Image analysis based on machine learning reliably identifies haematological malignancies challenging for human eye

Image analysis utilising neural networks can help identify details in tissue samples which are difficult to discern by the human eye. A study conducted at the University of Helsinki demonstrated that the technique makes it possible to accurately determine genetic mutations in the cancer cells of patients suffering from myelodysplastic syndrome, a malignant blood disorder.

Myelodysplastic syndrome (MDS) is a disease of the stem cells in the bone marrow, which disturbs the maturing and differentiation of blood cells. Annually, some 200 Finns are diagnosed with MDS, which can develop into acute leukaemia. Globally, the incidence of MDS is 4 cases per 100,000 person years.

To diagnose MDS, a bone marrow sample is needed to also investigate genetic changes in bone marrow cells. The syndrome is classified into groups to determine the nature of the disorder in more detail.

In the study conducted at the University of Helsinki, microscopic images of MDS patients’ bone marrow samples were examined utilising an image analysis technique based on machine learning. The samples were stained with haematoxylin and eosin (H&E staining), a procedure that is part of the routine diagnostics for the disease. The slides were digitised and analysed with the help of computational deep learning models.

The study was published in the Blood Cancer Discovery, a journal of the American Association for Cancer Research, and the results can also be explored with an interactive tool.

By employing machine learning, the digital image dataset could be analysed to accurately identify the most common genetic mutations affecting the progression of the syndrome, such as acquired mutations and chromosomal aberrations. The higher the number of aberrant cells in the samples, the higher the reliability of the results generated by the prognostic models.

Diagnosis supported by data analysis

One of the greatest challenges of utilising neural network models is understanding the criteria on which they base their conclusions drawn from data, such as information contained in images. The recently released study succeeded in determining what deep learning models see in tissue samples when they have been taught to look for, for example, genetic mutations related to MDS. The technique provides new information on the effects of complex diseases on bone marrow cells and the surrounding tissues.

“The study confirms that computational analysis helps to identify features that elude the human eye. Moreover, data analysis helps to collect quantitative data on cellular changes and their relevance to the patient’s prognosis,” says Professor Satu Mustjoki.

Digitoituja ja värjättyjä kudosnäytteitä

Part of the analytics carried out in the study was implemented using the Helsinki University Hospital (HUS) data lake environment, which enables the efficient collection and analysis of extensive clinical datasets.

“We’ve developed solutions to structure and analyse data stored in the HUS data lake. Image analysis helps us analyse large quantities of biopsies and rapidly produce diverse information on disease progression. The techniques developed in the project are suited to other projects as well, and they are perfect examples of the digitalizing medical science,” says doctoral student Oscar Brück.

The study received funding from the Cancer Foundation and the Sigrid Jusélius Foundation, as well as state funding for university-level health research (VTR). The study was carried out under the iCAN Digital Precision Cancer Medicine Flagship funded by the Academy of Finland.

/Public Release. This material comes from the originating organization and may be of a point-in-time nature, edited for clarity, style and length. View in full here.