An interpretable machine learning system for colorectal cancer diagnosis from pathology slides.

Considering the profound transformation affecting pathology practice, we aimed to develop a scalable artificial intelligence (AI) system to diagnose colorectal cancer from whole-slide images (WSI). For this, we propose a deep learning (DL) system that learns from weak labels, a sampling strategy that reduces the number of training samples by a factor of six without compromising performance, an approach to leverage a small subset of fully annotated samples, and a prototype with explainable predictions, active learning features and parallelisation. Noting some problems in the literature, this study is conducted with one of the largest WSI colorectal samples dataset with approximately 10,500 WSIs. Of these samples, 900 are testing samples. Furthermore, the robustness of the proposed method is assessed with two additional external datasets (TCGA and PAIP) and a dataset of samples collected directly from the proposed prototype. Our proposed method predicts, for the patch-based tiles, a class based on the severity of the dysplasia and uses that information to classify the whole slide. It is trained with an interpretable mixed-supervision scheme to leverage the domain knowledge introduced by pathologists through spatial annotations. The mixed-supervision scheme allowed for an intelligent sampling strategy effectively evaluated in several different scenarios without compromising the performance. On the internal dataset, the method shows an accuracy of 93.44% and a sensitivity between positive (low-grade and high-grade dysplasia) and non-neoplastic samples of 0.996. On the external test samples varied with TCGA being the most challenging dataset with an overall accuracy of 84.91% and a sensitivity of 0.996.

Pathologist Computer-Aided Diagnostic Scoring of Tumor Cell Fraction: A Swiss National Study.

Tumor cell fraction (TCF) estimation is a common clinical task with well-established large interobserver variability. It thus provides an ideal test bed to evaluate potential impacts of employing a tumor cell fraction computer-aided diagnostic (TCFCAD) tool to support pathologists' evaluation. During a National Slide Seminar event, pathologists (n = 69) were asked to visually estimate TCF in 10 regions of interest (ROIs) from hematoxylin and eosin colorectal cancer images intentionally curated for diverse tissue compositions, cellularity, and stain intensities. Next, they re-evaluated the same ROIs while being provided a TCFCAD-created overlay highlighting predicted tumor vs nontumor cells, together with the corresponding TCF percentage. Participants also reported confidence levels in their assessments using a 5-tier scale, indicating no confidence to high confidence, respectively. The TCF ground truth (GT) was defined by manual cell-counting by experts. When assisted, interobserver variability significantly decreased, showing estimates converging to the GT. This improvement remained even when TCFCAD predictions deviated slightly from the GT. The standard deviation (SD) of the estimated TCF to the GT across ROIs was 9.9% vs 5.8% with TCFCAD (P < .0001). The intraclass correlation coefficient increased from 0.8 to 0.93 (95% CI, 0.65-0.93 vs 0.86-0.98), and pathologists stated feeling more confident when aided (3.67 ± 0.81 vs 4.17 ± 0.82 with the computer-aided diagnostic [CAD] tool). TCFCAD estimation support demonstrated improved scoring accuracy, interpathologist agreement, and scoring confidence. Interestingly, pathologists also expressed more willingness to use such a CAD tool at the end of the survey, highlighting the importance of training/education to increase adoption of CAD systems.

Evaluation of MTAP and p16 immunohistochemical deficiency as surrogate marker for CDKN2A/B homozygous deletion in gliomas.

Homozygous deletion (HD) of the CDKN2A/B locus has emerged as an unfavourable prognostic marker in diffuse gliomas, both IDH-mutant and IDH-wild-type. Testing for CDKN2A/B deletions can be performed by a variety of approaches, including copy number variation (CNV) analysis based on gene array analysis, next generation sequencing (NGS) or fluorescence in situ hybridisation (FISH), but questions remain regarding the accuracy of testing modalities. In this study, we assessed: (1) the utility of S-methyl-5'-thioadenosine phosphorylase (MTAP) and cellular tumour suppressor protein pl61NK4a (p16) immunostainings as surrogate markers for CDKN2A/B HD in gliomas, and (2) the prognostic value of MTAP, across different histological tumour grades and IDH mutation status. One hundred consecutive cases of diffuse and circumscribed gliomas (Cohort 1) were collected, in order to correlate MTAP and p16 expression with the CDKN2A/B status in the CNV plot of each tumour. IDH1 R132H, ATRX and MTAP immunohistochemistry was performed on next generation tissue microarrays (ngTMAs) of 251 diffuse gliomas (Cohort 2) for implementing survival analysis. Complete loss of MTAP and p16 by immunohistochemistry was 100% and 90% sensitive as well as 97% and 89% specific for CDKN2A/B HD, respectively, as identified on CNV plot. Only two cases (2/100) with MTAP and p16 loss of expression did not demonstrate CDKN2A/B HD in CNV plot; however, FISH analysis confirmed the HD for CDKN2A/B. Moreover, MTAP deficiency was associated with shortened survival in IDH-mutant astrocytomas (n=75; median survival 61 vs 137 months; p<0.0001), IDH-mutant oligodendrogliomas (n=59; median survival 41 vs 147 months; p<0.0001) and IDH-wild-type gliomas (n=117; median survival 13 vs 16 months; p=0.011). In conclusion, MTAP immunostaining is an important complement for diagnostic work-up of gliomas, because of its excellent correlation with CDKN2A/B status, robustness, rapid turnaround time and low costs, and provides significant prognostic value in IDH-mutant astrocytomas and oligodendrogliomas, while p16 should be used cautiously.

Systematically higher Ki67 scores on core biopsy samples compared to corresponding resection specimen in breast cancer: a multi-operator and multi-institutional study.

Ki67 has potential clinical importance in breast cancer but has yet to see broad acceptance due to inter-laboratory variability. Here we tested an open source and calibrated automated digital image analysis (DIA) platform to: (i) investigate the comparability of Ki67 measurement across corresponding core biopsy and resection specimen cases, and (ii) assess section to section differences in Ki67 scoring. Two sets of 60 previously stained slides containing 30 core-cut biopsy and 30 corresponding resection specimens from 30 estrogen receptor-positive breast cancer patients were sent to 17 participating labs for automated assessment of average Ki67 expression. The blocks were centrally cut and immunohistochemically (IHC) stained for Ki67 (MIB-1 antibody). The QuPath platform was used to evaluate tumoral Ki67 expression. Calibration of the DIA method was performed as in published studies. A guideline for building an automated Ki67 scoring algorithm was sent to participating labs. Very high correlation and no systematic error (p = 0.08) was found between consecutive Ki67 IHC sections. Ki67 scores were higher for core biopsy slides compared to paired whole sections from resections (p ≤ 0.001; median difference: 5.31%). The systematic discrepancy between core biopsy and corresponding whole sections was likely due to pre-analytical factors (tissue handling, fixation). Therefore, Ki67 IHC should be tested on core biopsy samples to best reflect the biological status of the tumor.

Tumour budding/T cell infiltrates in colorectal cancer: proposal of a novel combined score.

AIMS: The tumour-node-metastasis classification system is used for prognostication purposes and to guide patient management. However, in colorectal cancer (CRC), additional markers are needed to stratify prognostic subgroups. Two promising markers have emerged from large bodies of research: tumour budding and T cell host response (CD3, CD8 and CD45RO infiltrates). However, attempts to combine these two parameters have been sparse. The aim of this study was to perform an assessment of potential protagonists that could be used in a combined score (budding/T cell score, BTS). METHODS AND RESULTS: This descriptive, retrospective study was performed on a multipunch tissue microarray containing material from 345 patients with stages I-IV CRC. Areas from tumour centre, front and microenvironment were stained for pancytokeratin/CD3, pancytokeratin/CD8 and pancytokeratin/CD45RO. Tumour buds were scored manually and T cell infiltrates digitally using open-source software. Tumour buds, T cell counts and combined BTS were associated with clinicopathological features and overall survival (OS). A higher combined BTS score (buds/CD8, tumour centre) performed better than budding or CD8/CD3 alone in predicting nodal metastases (P < 0.0001, OR = 1.466, 95% CI = 1.115-1.928). Only higher BTS (buds/CD3) were significantly associated with poorer OS on multivariate analysis (P = 0.012, hazard ratio = 1.218, 95% confidence interval = 1.044-1.419). CONCLUSIONS: Although CD8+ /CD3+ T cells are predictive of tumour biology in CRC, we found a combined BTS to be stronger in predicting survival and certain features with high clinical relevance, such as nodal metastases, in comparison to budding or T cells alone. Further studies combining T cell infiltrates and tumour budding are necessary to optimise risk assessment of CRC.