The growing adoption of deep learning (DL) in early-stage cancer diagnosis has demonstrated remarkable performance across multiple imaging tasks. Yet, the lack of transparency in these models (“black-box” problem) limits their adoption in clinical environments. This study proposes a methodological framework for developing interpretable DL models to support the early histopathological diagnosis of lung cancer, with a focus on adenocarcinoma and squamous cell carcinoma. The approach leverages publicly available datasets (TCGA-LUAD, TCGA-LUSC, LC25000) and employs high-performing architectures such as EfficientNet, along with post hoc explainability techniques including Grad-CAM and SHAP. Data will be pre-processed and sampled using stratified purposeful strategies to ensure diversity and balance across subtypes and stages. Model evaluation will combine standard performance metrics with clinician feedback and the spatial alignment of visual explanations with ground-truth annotations. While implementation remains a future step, this paper proposes a methodological framework designed to guide the development of DL systems that are not only accurate but also interpretable and clinically meaningful.

Background: Human papillomavirus (HPV) is diagnosed in most cervical cancers and is the fourth cause of cancer worldwide. Currently, fourteen HPV types are considered carcinogenic (HPV 16/18/31/33/35/39/45/51/52/56/58/59/66/68) with different oncogenicity. Cervical lesions might be due to HR (high risk)-HPV for what is important to identify new methods to genotype cervical lesions. The main goal was to identify, for the first time, HPV-HR in cervical lesions using Anyplex™ Detection kit in paraffin-embedded tissue. (2) Methods: It was performed a retrospective study in 45 women from Academic Hospital of Cova da Beira (CHUCB). DNA extraction was performed and through Anyplex™ II HPV-HR Detection kit HPV-HR were genotyped. (3) Results: HPV-HR were identified in 38 women; HPV-16 (55,26%), HPV-18/39/56/58/59 (5,26%), HPV-31 (21,05%), HPV-35 (7,89%), HPV-51/6 (2,63%) and HPV-52 (10,53%). (4) Conclusions: Our results reflect that Anyplex™ II HPV-HR Detection kit, designed for HR-HPV genotyping of cervical cancer screening, might be properly used to HR-HPV genotyping in histology. The proposed methodology allows an easy and cheaper technique that should be used as a future cervical risk stratification. It is a strong tool to be implemented in clinical practice in order to detect HR-HPV detection in cervical lesions, contributing to a more accurate diagnosis.

Background: Human papillomavirus (HPV), a leading cause of cervical cancer, is present in most cases of the disease and ranks as the fourth most common cancer in women globally. Among the HPV types, fourteen (HPV 16/18/31/33/35/39/45/51/52/56/58/59/66/68) are recognized as high-risk (hrHPV), each with varying levels of oncogenic potential. Detecting and genotyping these hrHPV types in cervical lesions is crucial, requiring the development of new diagnostic methods. Methods: This study focuses on a retrospective analysis conducted on 44 women from the Cova da Beira Local Health Unit. We used the Anyplex™ II hrHPV Detection kit for hrHPV genotyping from paraffin-embedded cervical tissue samples. Results: hrHPV types were identified in 38 out of the 44 women. Genotyping revealed HPV-16 (55.3%), HPV-18/39/56/58/59 (5.3%), HPV-31 (21.1%), HPV-35 (7.9%), HPV-51/66 (2.6%), and HPV-52 (10.5%). Conclusions: This study demonstrates that the Anyplex™ II hrHPV Detection kit, originally designed for cervical cancer screening, is also effective for hrHPV genotyping in histological analyses. This methodology offers a simpler and more cost-effective approach for cervical cancer risk stratification. Its implementation in clinical practice could enhance the detection of hrHPV in cervical lesions, thereby contributing to more precise diagnoses and potentially more informed treatment strategies.

Lung cancer is a leading cause of cancer-related deaths worldwide, and its diagnosis must be carried out as soon as possible to increase the survival rate. The development of computer-aided diagnosis systems can improve the accuracy of lung cancer diagnosis while reducing the workload of pathologists. The purpose of this study was to develop a learning algorithm (CancerDetecNN) to evaluate the presence or absence of tumor tissue in lung whole-slide images (WSIs) while reducing the computational cost. Three existing deep neural network models, including different versions of the CancerDetecNN algorithm, were trained and tested on datasets of tumor and non-tumor tiles extracted from lung WSIs. The fifth version of CancerDetecNN (CancerDetecNN Version 5) outperformed all existing convolutional neural network (CNN) models in the provided dataset, achieving higher precision (0.972), an area under the curve (AUC) of 0.923, and an F1-score of 0.897, while requiring 1 h and 51 min less for training than the best compared CNN model (ResNet-50). The results for CancerDetecNN Version 5 surpass the results of some architectures used in the literature, but the relatively small size and limited diversity of the dataset used in this study must be considered. This paper demonstrates the potential of CancerDetecNN Version 5 for improving lung cancer diagnosis since it is a dedicated model for lung cancer that leverages domain-specific knowledge and optimized architecture to capture unique characteristics and patterns in lung WSIs, potentially outperforming generic models in this domain and reducing the computational cost.

The immunohistochemical (IHC) expression of PD-L1 in cancer models is used as a predictive biomarker of response to immunotherapy. We aimed to evaluate the impact of the usage of 3 different tissue processors in the IHC expression of PD-L1 antibody clones: 22C3 and SP142. Three different topographies of samples (n = 73) were selected at the macroscopy room: 39 uterine leiomyomas, 17 placentas and 17 palatine tonsils. Three fragments were collected from each sample and were inked with a specific color that represented their separate processing in a different tissue processor (A, B or C). During embedding, the 3 fragments with distinct processing were ensemble in the same cassette for sectioning of 3 slides/each: hematoxylin-eosin, 22C3 PDL1 IHC staining and SP142 PD-L1 IHC staining, that were blindly observed by 2 pathologists under digital environment. All but one set of 3 fragments were considered adequate for observation even in the presence of artifacts associated with processing issues that were recorded as high as 50.7 % for processor C. The occurrence of background non-specific staining and the presence of false positive results appear to be unrelated with the PD-L1 clone or the type of tissue processing. 22C3 PD-L1 was more frequently considered adequate for evaluation than SP142 PD-L1 that, in 29.2 % of WSIs (after tissue processor C) were considered not adequate for observation due to lack of the typical pattern of expression. Similarly, the intensity of PD-L1 staining was significantly decreased in fragments processed by C (both PD-L1 clones) in tonsil and placenta specimens, and by A (both clones) in comparison with those processed by B. This study demonstrates the need to standardize the tissue processing in pathology to cope with the growing needs of precision medicine quantifications and the production of high-quality material necessary for computational pathology usage.

Paige Prostate is a clinical-grade artificial intelligence tool designed to assist the pathologist in detecting, grading, and quantifying prostate cancer. In this work, a cohort of 105 prostate core needle biopsies (CNBs) was evaluated through digital pathology. Then, we compared the diagnostic performance of four pathologists diagnosing prostatic CNB unaided and, in a second phase, assisted by Paige Prostate. In phase 1, pathologists had a diagnostic accuracy for prostate cancer of 95.00%, maintaining their performance in phase 2 (93.81%), with an intraobserver concordance rate between phases of 98.81%. In phase 2, pathologists reported atypical small acinar proliferation (ASAP) less often (about 30% less). Additionally, they requested significantly fewer immunohistochemistry (IHC) studies (about 20% less) and second opinions (about 40% less). The median time required for reading and reporting each slide was about 20% lower in phase 2, in both negative and cancer cases. Lastly, the average total agreement with the software performance was observed in about 70% of the cases, being significantly higher in negative cases (about 90%) than in cancer cases (about 30%). Most of the diagnostic discordances occurred in distinguishing negative cases with ASAP from small foci of well-differentiated (less than 1.5 mm) acinar adenocarcinoma. In conclusion, the synergic usage of Paige Prostate contributes to a significant decrease in IHC studies, second opinion requests, and time for reporting while maintaining highly accurate diagnostic standards.

Cell blocks may be hard to be totally automatically detected by the scanner (ADS), generating incomplete whole slide images (WSIs), with areas that are not scanned, leading to possible false negative diagnosis. The aim of this study is to test if inking the cell blocks helps increasing ADS. Test 1: 15 cell blocks were sectioned, one half inked black (1HB) and the other inked green (1HG). Each of the halves was individually processed to generate a WSI stained by the H&E. 1HBs and 1HGs had similar scanning time (median 59 s vs. 65 s, p = .126) and file sizes (median 382 Mb vs. 381 Mb, p = .567). The black ink interfered less in the observation (2.2% vs. 44.4%; p < .001) than in the green one. Test 2: 15 cell blocks were sectioned, one half inked black (2HB) and the other left unstained/null (2HN). Each of the halves was individually processed to generate three WSIs-one HE, one periodic-acid Schiff (PAS), and one immunostained by cytokeratin AE1&AE3 (CKAE1AE3). HE and PAS WSIs from both 2HN and 2HB groups were all totally ADS and had similar scanning times and file sizes. Concerning immunostaining with CKAE1AE3: ADS (46.7% vs. 93.3%; p = .014), median time for scanning (57 s vs. 83 s; p < .001) and file size (178 Mb vs. 338 Mb; p < .001) were reduced significantly in the 2HN group in comparison with the 2HB. Although increasing scanning time and file size, inking the cell blocks helps increasing ADS after immunostaining, improving the safety and efficiency of the digital pathology workflow.

The important developments achieved in recent years with a consequent paradigm shift in the treatment of non–small cell lung cancer (NSCLC), including the latest immune checkpoint inhibitors, have led to an increasing need to optimize the scarce material usually available in the diagnosis of these tumors. In this sense, this study intends to evaluate the performance of double immunohistochemistry (IHC) in comparison to simple IHC for programmed death-ligand 1 (PD-L1) evaluation with 22C3 clone for selection to therapy with pembrolizumab. For that, 38 histologic samples of NSCLC small biopsies sent to our laboratory were selected. Double IHC were performed with the doublets TTF1/PD-L1 and p40/PD-L1, after all the usual diagnostic routine and molecular study was performed. The slides were interpreted by 2 independent pathologists and the results obtained were compared with each other and with the results obtained at diagnosis. A perfect agreement was observed when comparing the immunoexpression of TTF1 and p40 in double IHC in relation to single IHC. Although the agreement was substantial in the analysis of the positive/negative PD-L1 IHC (81.6% to 92.1%; κ=0.610 to 0.829) and in the analysis of the 50% cut-off (86.8% to 89.5%; κ=0.704 to 0.759), it fell short of the expected and desirable agreement for a biomarker such as PD-L1, since this result will have a major role in the institution of a treatment. In conclusion, this small series does not allow us to recommend this methodology for the evaluation of the PD-L1 biomarker in double staining IHC with the 22C3 clone for therapy selection.

Lung cancer is the leading cause of cancer mortality worldwide, and it is urgently necessary to diagnose it as early as possible. Usually, the diagnostic process begins with a radiological examination which, when a possible tumour is present, is followed by a biopsy to extract tissue samples from the patient's lungs. Therefore, the purpose of this study is the development of an artificial intelligence algorithm that will analyse the Whole Slide Image (WSI) generated by the digitisation of the glass slides obtained from the extracted samples and detect if there is a tumour. The developed learning algorithms as well as the tested neural networks (NNs) were trained on a dataset composed of previously annotated WSI tiles, classified as Tumour or Non-Tumour. From these, four developed convolutional neural networks stood out and were selected to be compared with each other and with the tested NNs. When the best result of each of the developed architectures was compared to the highest result of the tested NNs, it was possible to denote that version 4 of CancerDetecNN achieved an average accuracy of 89.749 \% and an average loss of 0.220. Furthermore, the results for the four selected versions are in agreement with the results reported in the literature, however, the limited size of the given dataset must be considered. Given the results obtained, the fourth version has the potential to optimise the lung cancer diagnosis process.

Lung cancer is the type of cancer that causes most deaths worldwide and as sooner it is discovered as more possibilities there are for the patient to be treated. An accurate histological classification of tumours is essential for lung cancer diagnosis and adequate patient management. Whole-slide images (WSI) generated from tissue samples can be analysed using Deep Learning techniques to assist pathologists. In this study it is given an overview of the lung cancer exploring the different types of implementations undertaken until the present. These methods show a two-step implementation in which the tasks consist primarily of the detection of the tumour and after on the histologic classification of the tumour. To detect the neoplastic cells, the WSI is split in patches, and then a convolutional neural network is applied to identify and generate a heatmap highlighting the tumour regions. In the next step, features are extracted from the neoplasic regions and submitted in a classifier to determine the histologic type of tumour present in each patch. Moreover, in this paper, it is proposed a possible approach based on the literature review to surpass the limitations found in the actual models, and with better performance and accuracy, that could be used as an aid in the pathological diagnosis of the lung cancer.

In gastric cancer (GC), biomarkers that define prognosis and predict treatment response remain scarce. We hypothesized that the extent of CD44v6 membranous tumor expression could predict prognosis and therapy response in GC patients. Two GC surgical cohorts, from Portugal and South Korea (n = 964), were characterized for the extension of CD44v6 membranous immuno-expression, clinicopathological features, patient survival, and therapy response. The value of CD44v6 expression in predicting response to treatment and its impact on prognosis was determined. High CD44v6 expression was associated with invasive features (perineural invasion and depth of invasion) in both cohorts and with worse survival in the Portuguese GC cohort (HR 1.461; 95% confidence interval 1.002–2.131). Patients with high CD44v6 tumor expression benefited from conventional chemotherapy in addition to surgery (p < 0.05), particularly those with heterogeneous CD44v6-positive and -negative populations (CD44v6_3+) (p < 0.007 and p < 0.009). Our study is the first to identify CD44v6 high membranous expression as a potential predictive marker of response to conventional treatment, but it does not clarify CD44v6 prognostic value in GC. Importantly, our data support selection of GC patients with high CD44v6-expressing tumors for conventional chemotherapy in addition to surgery. These findings will allow better stratification of GC patients for treatment, potentially improving their overall survival.

The advantages of the digital methodology are well known. In this paper, we provide a detailed description of the process for the digital transformation of the pathology laboratory at IPATIMUP, the major modifications that operate throughout the processing pipeline, and the advantages of its implementation. The model of digital workflow implementation at IPATIMUP demonstrates that careful planning and adoption of simple measures related to time, space, and sample management can be adopted by any pathology laboratory to achieve higher quality and easy digital transformation.

Surgical resection with lymphadenectomy and perioperative chemotherapy is the universal mainstay for curative treatment of gastric cancer (GC) patients with locoregional disease. However, GC survival remains asymmetric in West‐ and East‐world regions. We hypothesize that this asymmetry derives from differential clinical management. Therefore, we collected chemo‐naïve GC patients from Portugal and South Korea to explore specific immunophenotypic profiles related to disease aggressiveness and clinicopathological factors potentially explaining associated overall survival (OS) differences. Clinicopathological and survival data were collected from chemo‐naïve surgical cohorts from Portugal (West‐Europe cohort [WE‐C]; n = 170) and South Korea (East‐Asia cohort [EA‐C]; n = 367) and correlated with immunohistochemical expression profiles of E‐cadherin and CD44v6 obtained from consecutive tissue microarrays sections. Survival analysis revealed a subset of 12.4% of WE‐C patients, whose tumors concomitantly express E‐cadherin_abnormal and CD44v6_very high, displaying extremely poor OS, even at TNM stages I and II. These WE‐C stage‐I and ‐II patients tumors were particularly aggressive compared to all others, invading deeper into the gastric wall (P = .032) and more often permeating the vasculature (P = .018) and nerves (P = .009). A similar immunophenotypic profile was found in 11.9% of EA‐C patients, but unrelated to survival. Tumours, from stage‐I and ‐II EA‐C patients, that display both biomarkers, also permeated more lymphatic vessels (P = .003), promoting lymph node (LN) metastasis (P = .019), being diagnosed on average 8 years earlier and submitted to more extensive LN dissection than WE‐C. Concomitant E‐cadherin_abnormal/CD44v6_very‐high expression predicts aggressiveness and poor survival of stage‐I and ‐II GC submitted to conservative lymphadenectomy.

To study the concordance between pathologists in the diagnosis of follicular patterned thyroid lesions using both digital and conventional optical settings. Five pathologists reviewed 50 hematoxylin and eosin-stained slides of follicular patterned thyroid lesions using both digital (the D-Sight 2.0 scanner and navigator viewer) and conventional optical instruments with washout interval time. The mean concordance rate with the ground truth (GT) was similar between conventional optical and digital observation (83.2 and 85.2%, respectively). The most frequent reason for diagnostic discordance with GT on both systems was the evaluation of nuclear features (69.1% for conventional optical observation and 59.4% for digital observation). The intraobserver diagnostic concordance mean was 86.8%. Time for digital observation (mean time per case = 2.9 ± 0.8 min) was higher than that for conventional optical observation (mean time per case = 2.0 ± 0.7 min). Interobserver correlation of measurements was higher in the digital observation than the conventional optical observation. Conventional optical and digital observation settings showed a comparable accuracy for the diagnosis of follicular patterned thyroid nodules, as well as substantial intraobserver agreement and a significant improvement in the reproducibility of the measurements that support the use of digital diagnosis in thyroid pathology. The origins underlying the variability of the diagnosis were the same in both conventional optical microscopy and digital pathology systems.

OBJECTIVES This study evaluated the usefulness of artificial intelligence (AI) algorithms as tools in improving the accuracy of histologic classification of breast tissue. METHODS Overall, 100 microscopic photographs (test A) and 152 regions of interest in whole-slide images (test B) of breast tissue were classified into 4 classes: normal, benign, carcinoma in situ (CIS), and invasive carcinoma. The accuracy of 4 pathologists and 3 pathology residents were evaluated without and with the assistance of algorithms. RESULTS In test A, algorithm A had accuracy of 0.87, with the lowest accuracy in the benign class (0.72). The observers had average accuracy of 0.80, and most clinically relevant discordances occurred in distinguishing benign from CIS (7.1% of classifications). With the assistance of algorithm A, the observers significantly increased their average accuracy to 0.88. In test B, algorithm B had accuracy of 0.49, with the lowest accuracy in the CIS class (0.06). The observers had average accuracy of 0.86, and most clinically relevant discordances occurred in distinguishing benign from CIS (6.3% of classifications). With the assistance of algorithm B, the observers maintained their average accuracy. CONCLUSIONS AI tools can increase the classification accuracy of pathologists in the setting of breast lesions.

The gastrointestinal and general symptoms as well as the arthralgia were the predominant manifestations. The majority of patients showed clinical and endoscopic improvement in response to treatment with trimethoprim-sulfamethoxazole. However, in these cases periodic acid-Schiff positive macrophages can remain for years. Thus, in the absence of clinical deterioration, the presence of these structures is not indicative ofactive disease.

Antecedentes. La enfermedad de Whipple es una enferme - dad infecciosa crónica rara y con un crecimiento multisis - témico. Aunque no exista una estimación válida de su real prevalencia, fueron descritos cerca de 1.000 casos. Objetivos. Descripción de las características demográficas, manifestacio - nes clínicas, resultados de laboratorio, endoscópicos y anato - mopatológicos, y tipo de tratamiento, duración y respuesta. Métodos. Basados en la histología duodenal, fueron iden - tificados doce casos de enfermedad de Whipple, entre 1997 y 2010, en el Centro Hospitalario de Vila Nova de Gaia. Resultados. Nueve (75%) de los enfermos eran del sexo mas - culino y la edad mediana al tiempo de diagnóstico era de 58 años. Todos los enfermos manifestaron por lo menos un síntoma gastrointestinal y general. Las artralgias fueron regis - tradas en 4 enfermos (33%) y ocurrieron en un promedio de 6 años antes de la aparición de los síntomas gastrointestinales y generales. Diez enfermos fueron sometidos a un tratamien - to inicial con trimetoprima-sulfametoxazol. La duración del tratamiento inicial fue de un año en 8 casos (80%). Siete enfermos (70%) presentaron una resolución sintomática en - tre los 3 y 6 meses de antibioterapia, acompañada de una mejoría histológica y endoscópica. En estos mismos enfermos se verificó la permanencia de macrófagos periodic acid-Schiff positivos en el examen histológico, aunque en número signifi - cativamente menor y con distribución más difusa. Los síntomas gastrointestinales predominaron en los casos de recidiva clínica. Conclusión. Los síntomas predominantes fueron las manifestaciones gastrointestinales, generales y articulares. La mayoría de los enfermos presentó mejoría clínica y endoscó - pica en respuesta al tratamiento con trimetoprima-sulfame - toxazol. Sin embargo, los macrófagos periodic acid-Schiff po - sitivos pueden permanecer durante años y por eso no indican enfermedad activa en ausencia de deterioro clínico.