13.01.2022: The mitotic count is an important microscopic parameter for assessing tumour proliferation in a patient. Despite its merits, however, the method has had one significant drawback: the results often differ depending on the person performing the examination. This can lead to an incorrect diagnosis and to an erroneous assessment of whether a tumour is malignant or not. With the help of artificial intelligence (deep learning), an international team of researchers led by Vetmeduni has managed to improve the current method and make it considerably more reliable. All 23 participants in the study showed a significant improvement in the accuracy and reproducibility of the mitotic count. This represents an important step towards improved tumour diagnosis with the possibility for cost benefits.

Mitosis is the phase of the cell cycle in which the genetic material is duplicated and the cell divides into two genetically identical daughter cells. The number of cells undergoing mitosis in a tissue sample is a measure of cell division activity and therefore an indicator of tumour malignancy. In pathological diagnostics, the level of mitotic activity is used to assess whether a tumour is malignant or benign. But despite being an important histological parameter for the evaluation of tumours, the mitotic count has one weak point, namely the human factor: pathologists must not only decide which region of the tissue sample to examine; they must also distinguish mitotic figures from other structures, including lookalikes. This can lead to incorrect assessments of the mitotic activity of the tumour.

High-performance algorithms based on deep learning

Recent advances in the field of artificial intelligence, especially through deep learning, have allowed the development of high-performance algorithms that may improve standardization of the mitotic count, as a recently published international study led by Vetmeduni shows. In the study, canine tissue samples were used to investigate how algorithmic predictions can assist pathologists in detecting mitotic hotspot locations and improving the differentiation of mitotic figures from other cells.

Computer assistance significantly improves diagnostic results

“Our results demonstrate that computer assistance using an accurate deep learning–based model is a promising method for improving the reproducibility and accuracy of mitotic counts in histological tumour sections,” says the study’s first author, Christof A. Bertram of the Institute of Pathology at Vetmeduni. Full computer assistance (assistance in the selection of regions of interest and the identification of mitotic figures) was superior to partial computer assistance limited to the selection of the regions of interest only.

More accurate and less expensive

According to Bertram, the study also shows that “computer-assisted mitotic counts may be a valuable method for standardization in future research studies and routine diagnostic tumour assessment using digital microscopy.” An accurate and reproducible cancer diagnosis is important for finding an appropriate therapy for tumour patients. Additionally, diagnostic laboratories could benefit from the improved work efficiency of the hybrid examination method – for example, due to the computer-assisted preselection of the regions of interest – which could also create cost advantages and help relieve the burden on the healthcare system.

Description: The heat map shows the density of mitotic figures found by the algorithm in the different tumor regions. Lighter areas represent higher mitotic activity. For the examination of the tumor, the pathologists should find the region with the highest mitotic activity and this task can be improved by the algorithmic support.