Artificial intelligence (AI) has emerged as a promising adjunct in surgical pathology, particularly in the diagnosis of prostate cancer, where variability in interpretation or missed cancer foci can significantly affect patient management. This review provides a concise, practice-oriented overview of the two Food and Drug Administration (FDA)-cleared AI tools for prostate biopsy interpretation: Paige Prostate Detect and Ibex Prostate Detect (formerly Galen Second Read). We examine their regulatory indications, diagnostic performance and integration requirements within digital pathology workflows. Emphasis is placed on real-world implementation considerations, including variation in technical inputs and the level at which data are analysed. We highlight less obvious risks, such as domain shift and the potential for inequitable performance in under-represented patient populations. Trade-offs between sensitivity and specificity, particularly in the context of AI-assisted pathologist assessments, are discussed using data from clinical validation studies. We also consider the variable impact of AI tools depending on the user's expertise, noting enhanced diagnostic consistency for general pathologists. By highlighting both the opportunities and limitations of integrating AI into routine practice, we aim to provide pathologists with a pragmatic understanding of how these systems may influence diagnostic workflows and to emphasise that FDA clearance must be complemented by local validation as well as ongoing performance monitoring to ensure safe and equitable deployment.

Artificial intelligence (AI) tools for digital pathology have crossed the threshold from pilot project to clinical deployment, with regulatory approvals, commercial products and prospective trials now accumulating across multiple tumour types. However, global-scale implementation demands careful evaluation of both opportunities and challenges. Traditional histopathology faces mounting pressures including interobserver variability and escalating workloads. At the same time, molecular diagnostics have become increasingly complex, while core workflows have remained fundamentally unchanged for decades. Digital pathology infrastructure and AI algorithms promise solutions through enhanced diagnostic accuracy, improved efficiency, democratised access to expertise and unprecedented research capabilities. AI has already demonstrated clinical value across multiple subspecialties, including prostate, breast and colorectal cancer, with applications spanning automated biomarker quantification, molecular subtype prediction and prognostic modelling. Systematic reviews also identify important limitations that temper enthusiasm for AI adoption. These include heterogeneity in study design, publication bias favouring positive results, limited generalisability across diverse populations and practice settings, concentration of research in narrow subspecialties and insufficient external validation. Critical implementation barriers include substantial infrastructure requirements, concerns around data privacy and security, gaps in regulatory oversight and the risk that algorithmic bias may exacerbate health disparities. Professional bodies advocate for responsible deployment guided by evidence-based standards. Key priorities include rigorous validation, workforce training, workflow integration and continuous performance monitoring. Making transformation equitable, safe and effective requires balancing innovation with accountability. AI should augment rather than replace pathologist expertise and be supported by robust governance frameworks that prioritise patient safety and diagnostic quality over commercial consideration.

B cell acute lymphoblastic leukaemia (B-ALL) infrequently presents with undetectable or very low peripheral blood (PB) lymphoblasts, which may cause diagnostic difficulties. We report the clinicopathological features of paediatric patients with B-ALL who presented with low (<5%; n=31) PB lymphoblasts by flow cytometry. 15 patients presented with <1% PB lymphoblasts (range: 0.0%-0.9%), a level potentially undetectable by manual smear review and standard 100-cell count. Compared with patients presenting with ≥5% PB lymphoblasts (n=45), those with <5% had more frequent musculoskeletal symptoms, less frequent lymphadenopathy and hepatosplenomegaly, higher platelet counts, lower white blood cell counts and lower lactate dehydrogenase. The presence of low PB lymphoblasts was associated with favourable cytogenetics (ETV6::RUNX1 or high hyperdiploidy without IKZF1 deletion) and less frequent central nervous system involvement. Aleukemic/low PB lymphoblast presentation of B-ALL is associated with unique clinicopathological features that overlap with non-malignant disorders. A high degree of clinical suspicion and careful review of PB flow cytometry data are required for timely diagnosis.

In clinical practice, thyroid function tests (TFTs) usually comprise thyroid-stimulating hormone (TSH) and free thyroxine (FT4), with free triiodothyronine (FT3) measured less frequently. We report a 76-year-old woman with autoimmune thyroiditis presenting with raised TSH and incongruously elevated FT4. The discordance between biochemical findings and clinical status prompted further investigations for assay interference. TFTs were assessed using four different immunoassay platforms and equilibrium dialysis. Additional studies included heterophilic antibody blocking and polyethylene glycol (PEG) precipitation. PEG precipitation demonstrated a high-molecular-weight interferent, consistent with thyroid hormone autoantibodies. FT4 measured by two-step immunoassays (Abbott Alinity and Beckman DxI) aligned with FT4 reported by equilibrium dialysis, whereas one-step assays (Roche Cobas and Siemens Atellica) reported spuriously elevated FT4. One-step immunoassays are particularly susceptible to interference from thyroid hormone autoantibodies, which may yield misleading results. Recognition of potential TFT assay interference and the use of alternative methodologies are essential for accurate diagnosis and appropriate patient management.

Transdifferentiation from follicular lymphoma (FL) to histiocytic/dendritic cell sarcoma (HDS) is rare and requires molecular confirmation of shared clonal origin. Targetable mutations such as BRAF V600E may offer therapeutic opportunities in such aggressive neoplasms. We report an exceptional case of untreated localised FL transdifferentiated to an HDS after 18 years. Shared BCL2 rearrangement and mutation profile confirmed a clonal link, while the HDS acquired an additional BRAF V600E mutation. Treatment with BRAF/MEK inhibitors yielded a sustained 18-month clinical response. The disease later relapsed as high-grade B-cell lymphoma with MYC and BCL2 rearrangements (HGBCL-MYC/BCL2), still harbouring the BRAF mutation. Complete remission was achieved with Rituximab, Cyclophosphamide, Hydroxydaunorubicin, Oncovin and Prednisone, but the double-hit lymphoma relapsed 14 months later.This case illustrates sequential transformation from FL to BRAF-mutated HDS with excellent response to BRAF/MEK inhibition, followed by evolution into HGBCL-MYC/BCL2 responding transiently to immunochemotherapy, emphasising the value of repeated histological and molecular reassessment in FL evolution.

Triple-negative breast cancer (TNBC) is defined in routine practice by the absence of oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) expression using fixed immunohistochemical and in situ hybridisation cut-offs. Although this framework ensures reproducibility and regulatory consistency, it does not equate to biological uniformity. TNBC is an operational clinicopathological category defined by reproducible biomarker thresholds applied to continuous gradients of receptor expression. This review examines the evolution of hormone receptor assessment from semiquantitative composite scoring systems to the current ≥1% ER positivity threshold, and analyses the diagnostic and clinical implications of ER-low-positive (1%-10%) tumours. We discuss the rarity and interpretative challenges of ER-/PR+ phenotypes, the impact of evolving HER2 testing criteria, including borderline amplification and HER2-low categories, and the consequences of guideline variation on TNBC classification. We also address histological heterogeneity within TNBC and propose a pragmatic three-tier framework that distinguishes definite TNBC, context-dependent/borderline TNBC and non-TNBC categories. Collectively, these considerations highlight that TNBC is a regulatory definition anchored to diagnostic thresholds rather than a discrete molecular entity. Awareness of definitional sensitivity at receptor cut-off margins is essential for accurate reporting, avoidance of misclassification and informed multidisciplinary decision-making. Recognition of TNBC as a biologically heterogeneous spectrum has implications for therapeutic selection, trial eligibility and future refinement of classification systems.