This Week in Pathology — Aug 9, 2026

Welcome to This Week in Pathology. This week we're covering 10 notable papers spanning diagnostic criteria and classification challenges, the expanding role of genomic and molecular profiling in tissue and cytology specimens, and the limits of artificial intelligence and integrated diagnostics. Let's dive in.

We start with papers that ask whether our current diagnostic thresholds are drawn in the right place. In Virchows Archiv, a Dutch group makes the case that non-mucinous pulmonary lesions with adenocarcinoma in situ morphology can substantially exceed the three-centimetre World Health Organization size limit [1]. They completely embedded the affected lobe in two patients who presented with a pneumonia-like consolidation and ground-glass pattern on computed tomography — one hundred and thirty and three hundred and twelve tissue blocks respectively — and found pure lepidic growth with no invasion anywhere, in tumours measuring sixteen centimetres. They then went back to a nationwide cohort of pathological T3 node-negative adenocarcinomas larger than seven centimetres and found three of one hundred and twelve cases that were, on review, purely non-invasive lepidic proliferations, all with recurrence-free follow-up. The practical consequence is uncomfortable but clear: if you are going to call a large lepidic lesion invasive, that conclusion has to rest on adequate sampling, and for these cases that may mean sampling the entire lesion. Staying with diagnostic criteria, Modern Pathology reports the largest Asian cohort of spitzoid neoplasms to date — one hundred and forty lesions, mostly atypical Spitz tumours, with eight Spitz melanomas and six conventional melanomas with spitzoid features [2]. Malignant behaviour tracked with asymmetry, deep atypical mitoses, sheet-like growth, diffuse PRAME positivity and loss of p16, which was absent in roughly two-thirds of malignant cases versus fewer than one in ten atypical Spitz tumours. Their most usable contribution is a refined fluorescence in situ hybridisation threshold: rather than relying on any single abnormality, requiring at least two aberrations gave complete sensitivity and about ninety-two percent specificity for predicting nodal metastasis or local recurrence. Isolated single changes, particularly MYB loss, were common in indolent lesions and should not be over-read. Sequencing found initiating kinase fusions or HRAS mutations in about ninety percent of true Spitz neoplasms, including five previously undescribed fusions involving ROS1, RET, ALK and NTRK1, while the spitzoid mimics lacked fusions entirely and instead carried conventional melanoma drivers. Malignant Spitz melanomas invariably had secondary hits in CDKN2A/B, TP53 or CDK4 — supporting a stepwise progression model you can actually use at the bench.

The second theme is genomics extending what we can extract from routine specimens. In Cancer Cytopathology, a Japanese group asked whether residual liquid-based cytology suspensions — the ordinary leftovers from a routine cytology preparation — can go straight into the Lung Cancer Compact Panel without dedicated molecular preservation [5]. Across sixty-nine specimens, DNA integrity was generally well preserved while RNA integrity was highly variable, and CytoRich Red fixation outperformed ThinPrep. Every sample passed the DNA-based assay, but about one in eight failed the RNA module, driven mainly by low RNA yield. Among evaluable cases, concordance with companion diagnostics was close to ninety-six percent with sensitivity around ninety-two percent, falling to about eighty-nine percent if RNA failures are counted as negatives. The message for cytology laboratories is that this is feasible without special handling, provided you pay attention to fixative choice and cellularity — and that a negative fusion result on a low-yield sample deserves scepticism. In the Journal of Pathology, optical genome mapping was applied to fifty-one children with B-cell acute lymphoblastic leukaemia and compared against karyotyping, fluorescence in situ hybridisation, digital multiplex ligation-dependent probe amplification and targeted RNA sequencing [6]. Optical genome mapping agreed closely with conventional cytogenetics while uncovering complex rearrangements missed by standard methods in fifteen percent of cases — including a complex KMT2A fusion invisible to fluorescence in situ hybridisation — and altered genetic classification or risk stratification in one in ten patients. Two interpretive refinements mattered: normalising copy number values by DNA index improved concordance in near-triploid and near-tetraploid cases, and a new approach to the peripheral fringe zones of called variants improved agreement with digital multiplex ligation-dependent probe amplification. Notably, the methods were complementary rather than hierarchical — some fusions in complex ETV6::RUNX1 translocations were missed by RNA sequencing and needed fluorescence in situ hybridisation to confirm.

Rounding out the molecular theme, the American Journal of Clinical Pathology reports the largest single sequencing cohort of anaplastic thyroid carcinoma assembled so far — seven hundred and thirteen tumours from a commercial genomic database [8]. Two hundred and fifty genes carried pathogenic alterations, one hundred and eight of which had not previously been reported in this disease, with MTAP the most frequent at just over fifteen percent and co-deleted with CDKN2A/B more often than earlier work suggested. Almost every single case — over ninety-nine percent — harboured at least one alteration linked to a potentially actionable therapy. That is a hypothesis-generating figure rather than a promise of response, but for a tumour with so few options it argues strongly for comprehensive sequencing at diagnosis. Modern Pathology adds a transcriptomic counterpart in classic Hodgkin lymphoma, profiling twenty-five cases with a targeted immune-oncology panel [10]. Immunotherapy non-responders showed a proliferative, cytokine-activated but immunologically cold profile, with activation of E2F, G2M, MYC and the PI3K/AKT/mTOR axis and negative enrichment of inflammatory and IL6/JAK/STAT3 signalling, whereas responders showed an inflamed microenvironment phenotype. With only five non-responders, this is preliminary — but the direction of travel, an oncogene-driven versus inflamed dichotomy independent of prior treatment, is worth watching.

The third theme is where confidence in a diagnosis breaks down. Cancer Cytopathology delivers a useful reality check on artificial intelligence in cervical cytology [3]. Six convolutional neural network architectures trained to separate negative cases from high-grade squamous intraepithelial lesion on isolated, non-overlapping single cells performed superbly in that setting, with areas under the curve from about zero point nine five to zero point nine nine six. Applied without retraining to hyperchromatic crowded cell groups — precisely the pattern that trips up human screeners — performance collapsed, with areas under the curve between roughly zero point three nine and zero point six eight. Even the best architecture managed only zero point six eight. If you are evaluating a commercial cervical cytology algorithm, ask specifically how it was validated on crowded groups, because single-cell benchmarks do not transfer. Complementing this, Virchows Archiv reviewed three hundred and eighty-one tumours sequenced through a precision genomics programme and found histology–genomics discrepancies in nine cases, about two point four percent, driven by non-specific morphology and inconclusive or frankly misleading immunophenotypes [4]. Low frequency, but concentrated in the hardest cases — an argument for integrated sign-out rather than sequential reporting. Two further papers map difficult differentials: a systematic review in Virchows Archiv on distinguishing blastic plasmacytoid dendritic cell neoplasm from blastoid lymphoid and myeloid mimics and from myeloid malignancies expressing plasmacytoid dendritic cell markers [7], and a Human Pathology series of thirty-four SMARCA4-altered malignancies showing that on cytology these tumours have no consistent morphology, are mostly adenocarcinomas rather than undifferentiated tumours, carry heavy co-mutation burden with TP53, KRAS and STK11, and had a two-year overall survival of around forty-nine percent [9]. The practical caution there: SMARCA4 alteration is a prognostic marker, not a synonym for undifferentiated tumour.

If you only have time for one paper this week, make it the cervical cytology artificial intelligence study in Cancer Cytopathology [3]. It quantifies exactly how badly published deep learning performance can overstate real-world capability, and that should shape how every laboratory interrogates vendor validation data.

Here are the key takeaways from this week in Pathology. First, large lepidic pulmonary lesions presenting as pneumonia-like consolidation may be genuinely non-invasive, so a diagnosis of invasion in these cases demands exhaustive sampling. Second, in spitzoid neoplasms, do not over-call a single fluorescence in situ hybridisation abnormality — require at least two, and add sequencing, since fusion status separates true Spitz lesions from conventional melanoma mimics. Third, routine liquid-based cytology residue can support comprehensive lung panel testing, but fixative and cellularity determine whether the RNA fusion module succeeds. Fourth, optical genome mapping adds clinically relevant information beyond karyotype and fluorescence in situ hybridisation in paediatric B-cell acute lymphoblastic leukaemia, changing risk stratification in roughly one in ten children, while remaining complementary to existing assays. And fifth, artificial intelligence performance figures are only as good as the morphology they were tested on — demand validation on crowded, overlapping cell groups before trusting a screening algorithm.

That's your roundup for This Week in Pathology. The full transcript and references are available on the episode page in your AudioScholar library. This is an AI-curated summary — for clinical decisions, always consult primary sources and current guidelines. See you next week.

If this weekly briefing is useful, follow the show in your podcast app so new episodes arrive automatically. And if you have a paper you have been meaning to read, upload the PDF, or paste any link, at audioscholar dot C C. We will turn it into audio like this one, in any of thirty-one languages.

This is an automated summary generated by artificial intelligence, which can make mistakes. Always review the original source materials.