
At this time - overly detailed/confusing outputs from LLMs aside - it is also unlikely that existing ML/AI digital diagnostics can handle the communication aspects in a case such as this - Thanks to all for engaging!
✓Attending Urologic Pathologist 🔬 #gupath #prostatecancer Dept Pathology & Lab Med @MSKPathology opinions=mine & do not reflect the views of my employer

At this time - overly detailed/confusing outputs from LLMs aside - it is also unlikely that existing ML/AI digital diagnostics can handle the communication aspects in a case such as this - Thanks to all for engaging!

#5 Would ML/AI help with Grading & Quantification in a focus like this? - AI depends on the annotations models were trained with - threshold/watershed areas of morphology and minute foci could be expected to be as difficult for an AI product as for pathologists (IMO)

IMO, we might be better off with re-framing as “those with expertise” who due to the interpretive and communication aspects detailed above may differ in their opinions (as we do)… Which brings us to #AI

#4 Moving Beyond the Era of the Sage on the Stage As @smlungpathguy highlights, we (and our patients) are often confronted with the challenge of multiple differing opinions from “experts” - indeed, we have opinions on this thread from a veritable all star team of GU pathologists

Hence, regardless of how one assigns GS/GG, the communication is crucial and in such patient scenarios (this is the only carcinoma focus, no significant MRI lesions, etc) - risk stratification should not rely on GS/GG alone @VickersBiostats @BehfarEhdaieMD

This may also have relevance to selection for active surveillance as @uroegg @dr_coops @GladellPaner suggest: pubmed.ncbi.nlm.nih.gov/3007… pubmed.ncbi.nlm.nih.gov/4283… @lwdean19 @SeanFletcherMD

pubmed.ncbi.nlm.nih.gov/4186… In a cancer focus such as this one, the relevant question clinically may NOT be: does the patient have GG2 with 40% pattern 4 v. GG3 with 60% pattern 4 or even 😱 GG4, but rather, that the patient only has 1mm of pattern 4 (at most) @NickPickersgill

#3 Clinical Interpretation This is important because of recent data, including from our group at @MSKCancerCenter, demonstrating that quantity of pattern 4 - specifically, a calculated value of mm pattern 4 for the case - has important predictive value in GG2-4 prostate cancer

This note should convey the grading assessment challenge in a way that can be clearly understood My practice: “Accurate assessment of pattern 4 quantification is challenging due to the small size of the carcinoma focus” - points to quantity rather than grade…

#2 Reporting Pathologists debate whether to even assign a GS/GrdGrp to foci such as these…which is a balance between letting clinical colleagues know that this is challenging and providing a data element that can be used for management ** include a note in the report

A portion of this focus appears = small discrete glands on at least some of the levels provided (IMO), so that would leave some combination of pattern 3 and 4 and makes relevant the discussion points raised by @slusagar @GladellPaner and others

And then we are left with a threshold/watershed question: what is a small (?minute) discrete gland v. a poorly formed gland The latter is a particularly thorny area (agreement-wise) for even GU pathologists: pubmed.ncbi.nlm.nih.gov/2609… Next step: examine multiple tissue levels…

A portion of the focus = poorly formed glands with attenuated cytoplasm/lumens (IMO) - seeing this glandular structure prevents me from assessing any of this as pattern 5 (single cells with vacuoles) - anecdotally, this is a challenging area for AI models So, pattern 3 v. 4…

Depending on how one interprets the above - one could diagnose: majority small discrete glands/no single cells (i.e., 3+4), majority poorly formed glands/some small discrete/no single cells (i.e., 4+3), all PFG (i.e., 4+4), +/- some degree of single cells (i.e., GG5) What Now?

Simple… Discrete glands: pattern 3 Poorly formed glands: pattern 4 Single cells: pattern 5 More Complex… A focus like ours displaying: small glands (challenge: discrete v. poorly formed) + glands/cells with small lumens/vacuoles (challenge: poorly formed v. single cells)
