BACKGROUND:Liquid biopsy has been developed as an alternative to tissue-based sequencing for detecting genomic alterations in solid tumors. However, the clinical utility of liquid biopsy in patients with solid tumors for whom tissue-based next-generation sequencing (NGS) is infeasible has not been well-characterized, particularly in previously untreated individuals. METHODS:This prospective study evaluated the clinical impact of liquid biopsy, focusing on six solid tumor types. Overall, 109 patients were enrolled and underwent liquid biopsy using Guardant360 (Guardant Health, Redwood City, CA, USA). Among these, 94 (86.3%) patients were previously untreated. RESULTS:The most common cancer type was non-small cell lung cancer (n = 57, 52.3%), followed by pancreatic (n = 35, 32.1%), biliary tract (n = 8, 7.3%), gastric (n = 5, 4.6%), colorectal (n = 3, 2.8%), and triple-negative breast (n = 1, 0.9%) cancers. The success rate of liquid biopsy was 99.1%, and the median turnaround time from blood collection to results was 7 days (range: 5-22 days). Actionable alterations were detected in 31 (28.4%) patients, and 8.3% of them received matched therapy based on alterations identified by liquid biopsy. Among previously untreated patients, actionable mutations were identified in 29.8%, and 8.5% received matched therapy. CONCLUSIONS:In patients with advanced solid tumors for which tissue-based NGS is not feasible, performing upfront liquid biopsy could lead to the detection of actionable alterations and help guide targeted therapies. CLINICAL TRIAL REGISTRY:UMIN Clinical Trials Registry (UMIN000041722).

BACKGROUND:The detection of BRAF fusions by using next-generation sequencing (NGS) is essential for comprehensive analysis. METHODS:Data BRAF positive rearrangements from Chinese cancer patients were analyzed. DNA NGS was performed on FFPE samples, and RNA NGS was used to confirm fusion transcripts. RESULTS:BRAF fusions were identified in various cancers, predominantly glioma (87.8%). DNA NGS detected 371 BRAF fusion-positive samples, with 338 retaining the serine/threonine receptor tyrosine kinase domain (RTKD), divided into four groups: common (n = 254), rare (n = 66), intergenic (n = 7), and exonic (n = 11) fusions. Common fusions, mainly KIAA1549-BRAF, comprised the majority, with variations at introns 8, 9, and 10. Rare fusions and intergenic/exonic breakpoints displayed diverse structural patterns. RNA NGS verified transcriptional consistency in most samples from common fusions. However, various outcomes at the RNA level were found in other groups, involving mechanisms like alternative splicing, antisense rearrangement, and frameshift rearrangement. Additionally, 33 fusions lacked the RTKD, demonstrating significant structural diversity. Furthermore, 22 novel fusions were identified, which were distributed in tongue cancer, liver cancer, lung cancer, melanoma, brain cancer, and colon cancer. CONCLUSIONS:Comprehensive molecular profiling and RNA sequencing are essential for accurate fusion detection, improving the design of NGS panels and aiding in the targeted therapy of BRAF fusion-positive cancers.

OBJECTIVE:There is limited evidence on the economic implications of assessing patients' access to personalised treatments through Comprehensive Genomic Profiling (CGP) and Molecular Tumour Board (MTB), prompting the need to analyse their impact on the cost of the cancer diagnostic journey (from hospital admission to MTB evaluation) and accessibility to personalised therapies. DESIGN:Retrospective observational cohort. SETTING:Patients discussed from April 2020 to September 2021 by the institutional MTB operating at Fondazione IRCCS Istituto Nazionale Tumori of Milan, an Italian centre of excellence in oncology pertaining to the national health system. PARTICIPANTS:676 patients focused on: non-small cell lung cancer (NSCLC), cholangiocarcinoma (CCA), pancreatic carcinoma (PC) and gastro-oesophageal carcinoma (GEC). We defined two different scenarios: (1) patients tested with small Next-Generation Sequencing (NGS) panels (≤60 biomarkers) vs (2) patients tested with comprehensive panels (>60 biomarkers). MAIN OUTCOMES AND MEASURES:We measured (1) patients' eligibility to personalised therapies based on genomic data obtained using targeted somatic NGS panels, (2) MTB cost and the overall diagnostic journey cost and (3) the cost to find a patient eligible to access personalised treatments. RESULTS:Tumour profiling with comprehensive NGS panels improved patients' eligibility to personalised therapies compared with small panels (NSCLC: 39% comprehensive panel vs 37% small panel; CCA: 43% vs 17%; PC: 35% vs 3%; GEC: 40% vs 0%). The overall diagnostic journey cost per patient was between 3.2K and 7.4K (NSCLC: 7.4K comprehensive panel vs 6.4K small panel; CCA: 4.9K vs 3.7K; PC: 5.8K vs 4.5K; GEC: 4.2K vs 3.2K). MTB discussion accounted for only 2-3% of the diagnostic journey cost per patient (around 113€/patient). The cost to find patient eligible for personalised treatments varied significantly according to panel size and tumour setting (NSCLC: 5K comprehensive panel vs 2.8K small panel; CCA: 4.4K vs 4.4K; PC: 5.5K vs 27K; GEC: 5.2K vs not measurable since none of the patients analysed with small NGS panels were eligible). CONCLUSIONS AND RELEVANCE:MTB discussion of genomic data obtained with NGS comprehensive panels significantly increases patient eligibility to targeted therapies and optimise the cost to find a patient eligible to personalised treatments, mainly for CCA, PC and GEC patients.

The analysis of histopathology images with artificial intelligence aims to enable clinical decision support systems and precision medicine. The success of such applications depends on the ability to model the diverse patterns observed in pathology images. To this end, we present Virchow, the largest foundation model for computational pathology to date. In addition to the evaluation of biomarker prediction and cell identification, we demonstrate that a large foundation model enables pan-cancer detection, achieving 0.95 specimen-level area under the (receiver operating characteristic) curve across nine common and seven rare cancers. Furthermore, we show that with less training data, the pan-cancer detector built on Virchow can achieve similar performance to tissue-specific clinical-grade models in production and outperform them on some rare variants of cancer. Virchow's performance gains highlight the value of a foundation model and open possibilities for many high-impact applications with limited amounts of labeled training data.

The lack of annotated publicly available medical images is a major barrier for computational research and education innovations. At the same time, many de-identified images and much knowledge are shared by clinicians on public forums such as medical Twitter. Here we harness these crowd platforms to curate OpenPath, a large dataset of 208,414 pathology images paired with natural language descriptions. We demonstrate the value of this resource by developing pathology language-image pretraining (PLIP), a multimodal artificial intelligence with both image and text understanding, which is trained on OpenPath. PLIP achieves state-of-the-art performances for classifying new pathology images across four external datasets: for zero-shot classification, PLIP achieves F1 scores of 0.565-0.832 compared to F1 scores of 0.030-0.481 for previous contrastive language-image pretrained model. Training a simple supervised classifier on top of PLIP embeddings also achieves 2.5% improvement in F1 scores compared to using other supervised model embeddings. Moreover, PLIP enables users to retrieve similar cases by either image or natural language search, greatly facilitating knowledge sharing. Our approach demonstrates that publicly shared medical information is a tremendous resource that can be harnessed to develop medical artificial intelligence for enhancing diagnosis, knowledge sharing and education.

Much work has been published on artificial intelligence (AI) and oncology, with many focusing on an algorithm perspective. However, very few perspective articles have explicitly discussed the role of AI in oncology from the perspectives of the stakeholders-the clinicians and the patients. In this article, we delve into the opportunities of AI in oncology from the clinician's and patient's lens. From the clinician's perspective, we discuss reducing burnout, enhancing decision making, and leveraging vast data sets to provide evidence-based recommendations, eventually affecting diagnostic accuracy and treatment planning. From the patient's perspective, we discuss AI virtual concierge, which could improve the cancer care journey by facilitating patient education, mental health support, and personalized lifestyle wellness recommendations promoting a holistic approach to care. We aim to highlight the stakeholders' unmet needs and guide institutions to create innovative AI solutions in oncology. By addressing these perspectives, our article aims to bridge the gap between technological research advancements and their real-world AI-focused clinical applications in cancer care. Understanding and prioritizing the needs of the stakeholders will foster the development of impactful AI tools and intentional utilization of such technology, with an aim for clinical implementation and integration into workflows.

The goal of the methylation classifier in brain tumor classification is to accurately classify tumors based on their methylation profiles. Accurate brain tumor diagnosis is the first step for healthcare professionals to predict tumor prognosis and establish personalized treatment plans for patients. The methylation classifier can be used to perform classification on tumor samples with diagnostic difficulties due to ambiguous histology or mismatch between histopathology and molecular signatures, i.e., not otherwise specified (NOS) cases or not elsewhere classified (NEC) cases, aiding in pathological decision-making. Here, the authors elucidate upon the application of a methylation classifier as a tool to mitigate the inherent complexities associated with the pathological evaluation of brain tumors, even when pathologists are experts in histopathological diagnosis and have access to enough molecular genetic information. Also, it should be emphasized that methylome cannot classify all types of brain tumors, and it often produces erroneous matches even with high matching scores, so, excessive trust is prohibited. The primary issue is the considerable difficulty in obtaining reference data regarding the methylation profile of each type of brain tumor. This challenge is further amplified when dealing with recently identified novel types or subtypes of brain tumors, as such data are not readily accessible through open databases or authors of publications. An additional obstacle arises from the fact that methylation classifiers are primarily research-based, leading to the unavailability of charging patients. It is important to note that the application of methylation classifiers may require specialized laboratory techniques and expertise in DNA methylation analysis.

Sarcoma classification is challenging and can lead to treatment delays. Previous studies used DNA aberrations and machine-learning classifiers based on methylation profiles for diagnosis. We aimed to classify sarcomas by analyzing methylation signatures obtained from low-coverage whole-genome sequencing, which also identifies copy-number alterations. DNA was extracted from 23 suspected sarcoma samples and sequenced on an Oxford Nanopore sequencer. The methylation-based classifier, applied in the nanoDx pipeline, was customized using a reference set based on processed Illumina-based methylation data. Classification analysis utilized the Random Forest algorithm and t-distributed stochastic neighbor embedding, while copy-number alterations were detected using a designated R package. Out of the 23 samples encompassing a restricted range of sarcoma types, 20 were successfully sequenced, but two did not contain tumor tissue, according to the pathologist. Among the 18 tumor samples, 14 were classified as reported in the pathology results. Four classifications were discordant with the pathological report, with one compatible and three showing discrepancies. Improving tissue handling, DNA extraction methods, and detecting point mutations and translocations could enhance accuracy. We envision that rapid, accurate, point-of-care sarcoma classification using nanopore sequencing could be achieved through additional validation in a diverse tumor cohort and the integration of methylation-based classification and other DNA aberrations.