The evaluation of the number of mouse ovarian primordial follicles (PMF) can provide important information about ovarian function, regulation of folliculogenesis or the impact of chemotherapy on fertility. This counting, usually performed by specialized operators, is a tedious, time-consuming but indispensable procedure.The development and increasing use of deep machine learning algorithms promise to speed up and improve this process. Here, we present a new methodology of automatically detecting and counting PMF, using convolutional neural networks driven by labelled datasets and a sliding window algorithm to select test data. Trained from a database of 9 millions of images extracted from mouse ovaries, and tested over two ovaries (3 millions of images to classify and 2 000 follicles to detect), the algorithm processes the digitized histological slides of a completed ovary in less than one minute, dividing the usual processing time by a factor of about 30. It also outperforms the measurements made by a pathologist through optical detection. Its ability to correct label errors enables conducting an active learning process with the operator, improving the overall counting iteratively. These results could be suitable to adapt the methodology to the human ovarian follicles by transfer learning.

The origins of Big Data date back to 1941, when the first references were made to the notion of "information explosion" in the Oxford Dictionary of English. James Maar has highlighted in 1996 in a report of the National Academy of Sciences the concept of "massive data set" (1). But it was only in 1997 that the precise term 'Big Data' first appeared in an article in the Digital Library of the Association for Computing Machinery (2), referring to the technical challenge of analyzing large sets of data. It has since been used to designate "structured or unstructured data, whose very large volume requires adapted analysis tools". Web giants (Google, Amazon, Facebook, Apple, Twitter) have developed such tools over the past decade, ensuring a constant marginal cost of data exploitation, regardless of volume.

Background uPA and PAI-1 are breast cancer biomarkers that evaluate the benefit of chemotherapy (CT) for HER2-negative, estrogen receptor-positive, low or intermediate grade patients. Our objectives were to observe clinical routine use of uPA/PAI-1 and to build a new therapeutic decision tree integrating uPA/PAI-1. Methods We observed the concordance between CT indications proposed by a canonical decision tree representative of French practices (not including uPA/PAI-1) and actual CT prescriptions decided by a medical board which included uPA/PAI-1. We used a method of machine learning for the analysis of concordant and non-concordant CT prescriptions to generate a novel scheme for CT indications. Results We observed a concordance rate of 71% between indications proposed by the canonical decision tree and actual prescriptions. Discrepancies were due to CT contraindications, high tumor grade and uPA/PAI-1 level. Altogether, uPA/PAI-1 were a decisive factor for the final decision in 17% of cases by avoiding CT prescription in two-thirds of cases and inducing CT in other cases. Remarkably, we noted that in routine practice, elevated uPA/PAI-1 levels seem not to be considered as a sufficient indication for CT for N≤3, Ki 67≤30% tumors, but are considered in association with at least one additional marker such as Ki 67>14%, vascular invasion and ER-H score <150. Conclusions This study highlights that in the routine clinical practice uPA/PAI-1 are never used as the sole indication for CT. Combined with other routinely used biomarkers, uPA/PAI-1 present an added value to orientate the therapeutic choice.