DIGITAL PATHOLOGY
Automated slide QC Slide quality control is of the utmost importance in digital pathology workflows. Slide artefacts generated during the slide creation and scanning process, like dust, coverslip issues, pen marks, out-of-focus areas, and tissue folds, can all affect downstream analysis, regardless of whether it is being conducted by a human or an AI algorithm. A recent study including nearly 2,000 WSIs found that almost 10% of the digitised slides had artefacts that precluded subsequent computational analysis.11
Not only can artefacts
Fig 2. AI image analysis of PD-L1 on non-small cell lung cancer tissue. PD-L1 positive cells are shown in magenta, PD-L1 negative cells are shown in cyan, and other cells are shown in yellow.
enhanced patient care and decreased turnaround time through automation; which in turn provides uniform, reproducible results. AI can streamline workflows by automating labour-intensive, repetitive processes consistently and quantitatively.5,6
Additionally, AI-driven
analysis eliminates interobserver variability, ensuring precise and standardised results, which is crucial for reliable comparative studies.7
However, AI
is only as effective as the data it is trained on, necessitating human supervision to ensure the quality of algorithmic findings. AI is most effective when augmenting human capabilities, enhancing their efficiency and accuracy, rather than replacing them.
Regulating AI as a medical device The current regulatory landscape for AI tools is complex and rapidly evolving. In the EU, AI analysis algorithms for diagnostics are regulated as in-vitro devices, a system that has recently been updated to enhance the validation process for all approved devices. In the United States, AI tools for diagnostic purposes require 510(k) clearance through the Food and Drug Administration (FDA), with slight variations depending on whether their approval is predicated on a prior authorisation. While it is not currently possible to analyse the number of CE-IVD/CE-IVDR-marked AI products due to a lack of a mandated centralised database, the FDA has cleared only one AI algorithm for pathology diagnostics.
Identifying and grading malignancy AI’s most obvious application in digital pathology is identifying and grading malignancies. Algorithms that perform these tasks are intended to provide the reading pathologist with information that they can use to direct their diagnostic
46
process. While many validation studies have found that AI algorithms that identify and grade cancer capture small foci of neoplasia that are missed by human pathologists, it is also true that, in general, these algorithms can be confused by common processing artefacts and tissues that share morphological features, resulting in false positives.8,9
Algorithms
that scan slides for cancerous lesions can also be used to triage cases so that pathologists review the most urgent cases first, which can lead to improved patient care as those who need treatment most urgently are moved to the front of the line. AI algorithms can also be used as a post-diagnosis quality control step, which could reduce the necessity of second opinions.10
Grading cancerous lesions is another application AI is well-suited to assist with as it can provide a standardised and truly quantitative assessment of lesions in place of the semi-quantitative methods currently employed, which are subject to significant inter- and intra-observer variability.7,8
While truly
accurate comparative studies require that practices adopt the same algorithm, the standardisation of a highly variable process is extremely valuable, especially when nearly all aspects of treatment and prognostication depend on these calls being made accurately.
AI is most effective when augmenting human capabilities, enhancing their efficiency and accuracy, rather than replacing them
negatively affect the process of evaluating a slide, they can also directly prevent AI tools from working properly, resulting in errors. The process of digitisation and deploying AI-based automated artefact detection algorithms introduces an opportunity to simultaneously scan each slide for artefacts (Fig 1). In comparison to manual artefact screening, where only a small selection of slides is scanned from relatively low power, an automated AI tool for quality control can scan every slide completely, identifying those affected by artefacts and flagging them for recut and rescanning.
The same algorithms can identify artefact areas and annotate them, removing them from downstream processing and preventing those areas from being evaluated by AI algorithms while allowing the annotated high-quality tissue within the WSI to advance to downstream analysis. This enables slides that are affected by artefacts to still be diagnostically valuable and is especially useful in cases with small specimens or limited tissue to analyse, where a recut may not be possible. Slide quality control algorithms act as a filter for the pathology laboratory, enabling laboratory staff to have fine-grained control over the quality-control process. This results in faster turnaround times, increased staff availability, and higher quality slides for diagnostics, research, and archiving.
Evaluating IHC with AI IHC is an important tool that provides pathologists and other clinicians with information about the expression of specific biomarkers that can be used for diagnosis, treatment decision making, and prognostication. The degree to which certain biomarkers are expressed provides vital information about which therapies are likely to be effective for a given patient.
Despite this importance, the evaluation of IHC expression remains subject to a high degree of inter- and intra-observer variability, even with the introduction of semi-quantitative
FEBRUARY 2025
WWW.PATHOLOGYINPRACTICE.COM
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60