Automated Inmunohistochemical Staining Quantification In Human Biopsies: Preliminary Results Using Deep Learning

QUIÑONES M; DOCTOROVICH J; REVOLLO N; ALONSO E; FERNANDEZ CHAVES L; FACCHINETTI MM; CURINO AC; DELRIEUX C; COLÓ GP

Congreso; LXVI Reunión Anual de la Sociedad Argentina de Investigación Clínica (SAIC).; 2021

Sociedad Argentina de Investigación Clinica

Among the current challenges in histopathological assessmentfor diagnosis in clinical contexts is an accurate determination ofthe actual tissue malignancy. This task is often performed usingmicroscopy over immunohistochemical (IHQ) staining applied ontissue samples, on which several specialists judge the tissue condition following specific criteria. However, this task is proven to beprone to high inter- and intra-subject variance, which raises theneed to elaborate more robust tools and frameworks to assist onthis task. The recent influx of deep learning technologies, which areproven to be successful in a variety of contexts, appears to be anadequate alternative in this context. In this aim, we present a jointeffort between research groups from Cancer Biology Laboratory(INIBIBB-CONICET) and the Imaging Sciences Laboratory (LCIUNS-CONICET). Starting with IHQ stained images taken with Olympus CX31 microscope from thyroid and breast cancer biopsies, weapplied a Mask C-RNN network for cell nuclei detection. For thispurpose, we retrained the net with a series of labeled examples provided by the biochemical specialists. After this initial detection, a ROIwas determined surrounding the nuclei, within which the proportionof diaminobenzidine stain (brown-colored precipitation) is computedas a proxy indicator of the Immunoreactive Score (IRS). For this, aRandom Forest classifier was trained using stain/no stain labeledpixels also provided by the experts. The results appear promisingin the sense that the resulting system is able to consistently providemalignancy assessment even in difficult cases or when the quality ofthe microscopy acquisition is below standard