Programmed death-ligand 1 expression by digital image analysis advances thyroid cancer diagnosis among encapsulated follicular lesions

Abstract

Recognition of noninvasive follicular thyroid neoplasms with papillary-like nuclear features (NIFTP) that distinguishes them from invasive malignant encapsulated follicular variant of papillary thyroid carcinoma (EFVPTC) can prevent overtreatment of NIFTP patients. We and others have previously reported that programmed death-ligand 1 (PD-L1) is a useful biomarker in thyroid tumors; however, all reports to date have relied on manual scoring that is time consuming as well as subject to individual bias. Consequently, we developed a digital image analysis (DIA) protocol for cytoplasmic and membranous stain quantitation (ThyApp) and evaluated three tumor sampling methods [Systemic Uniform Random Sampling, hotspot nucleus, and hotspot nucleus/3,3′-Diaminobenzidine (DAB)]. A patient cohort of 153 cases consisting of 48 NIFTP, 44 EFVPTC, 26 benign nodules and 35 encapsulated follicular lesions/neoplasms with lymphocytic thyroiditis (LT) was studied. ThyApp quantitation of PD-L1 expression revealed a significant difference between invasive EFVPTC and NIFTP; but none between NIFTP and benign nodules. ThyApp integrated with hotspot nucleus tumor sampling method demonstrated to be most clinically relevant, consumed least processing time, and eliminated interobserver variance. In conclusion, the fully automatic DIA algorithm developed using a histomorphological approach objectively quantitated PD-L1 expression in encapsulated thyroid neoplasms and outperformed manual scoring in reproducibility and higher efficiency.

Impact of image analysis and artificial intelligence in thyroid pathology, with particular reference to cytological aspects

“…This finding is similar to that of another study that also showed that algorithm data are normally distributed, while manual scores are not . Other studies applied automated image analysis to evaluate the expression of various biomarkers including trophoblast antigen 2, programmed death‐ligand 1, monocarboxylate transporter 1/translocase of the outer mitochondrial membrane member 20, sodium‐iodine symporter, and CAP‐Gly domain‐containing linker protein 2 . Even with no comparison to visual assessment, it appears to be reliable to quantify immunohistochemical expression with image analysis.…”

“…In one study by this group, there was also a comparison between two AI models for immunohistochemistry assessment, in which the authors stressed that their algorithms provide results in the form of continuous data, that are easier to manipulate with statistics and to compare with results from other studies . This finding is similar to that of another study that also showed that algorithm data are normally distributed, while manual scores are not . Other studies applied automated image analysis to evaluate the expression of various biomarkers including trophoblast antigen 2, programmed death‐ligand 1, monocarboxylate transporter 1/translocase of the outer mitochondrial membrane member 20, sodium‐iodine symporter, and CAP‐Gly domain‐containing linker protein 2 .…”

“…A subset of studies compared the automated assessment of cytological and histological slides to manual assessment by a pathologist, reporting high correlation values between these two modalities with polyserial correlation coefficient, linear correlation coefficient, and area under the curve . In some studies, there was only a descriptive report of high correlation without measures .…”

“…The reported correlation coefficient with visual score ranged from 0.76 to 0.83 and the areas under the curve (AUCs) were 0.909‐0.932. In these publications, the investigators also reported that the AI application was faster and less time‐consuming than visual pathologist assessment …”

“…In most of the studies, automated algorithms were applied to quantify cellular parameters such as percentage of stained area, number of stained cells, intensity of staining for immunohistochemistry, and nuclear features in cytology studies . Three studies from a Belgian group explored the expression of various immunohistochemical markers in partially overlapping cohorts of cases with PTC, follicular variant of PTC (FVPTC), thyroid neoplasm with uncertain malignant potential, normal tissue and adenomatous lesions, both with or without semi‐quantitative assessment by the pathologist.…”