Citations

Abstract Background The radiation oncology clinical practice involves many steps relying on the dynamic interplay of abundant text data. Large language models have displayed remarkable capabilities in processing complex text information. But their direct applications in specific fields like radiation oncology remain underexplored. Purpose This study aims to investigate whether fine‐tuning LLMs with domain knowledge can improve the performance on Task (1) treatment regimen generation, Task (2) treatment modality selection (photon, proton, electron, or brachytherapy), and Task (3) ICD‐10 code prediction in radiation oncology. Methods Data for 15 724 patient cases were extracted. Cases where patients had a single diagnostic record, and a clearly identifiable primary treatment plan were selected for preprocessing and manual annotation to have 7903 cases of the patient diagnosis, treatment plan, treatment modality, and ICD‐10 code. Each case was used to construct a pair consisting of patient diagnostics details and an answer (treatment regimen, treatment modality, or ICD‐10 code, respectively) for the supervised fine‐tuning of these three tasks. Open source LLaMA2‐7B and Mistral‐7B models were utilized for the fine‐tuning with the Low‐Rank Approximations method. Accuracy and ROUGE‐1 score were reported for the fine‐tuned models and original models. Clinical evaluation was performed on Task (1) by radiation oncologists, while precision, recall, and F‐1 score were evaluated for Task (2) and (3). One‐sided Wilcoxon signed‐rank tests were used to statistically analyze the results. Results Fine‐tuned LLMs outperformed original LLMs across all tasks with p value ≤ 0.001. Clinical evaluation demonstrated that over 60% of the fine‐tuned LLMs‐generated treatment regimens were clinically acceptable. Precision, recall, and F1‐score showed improved performance of fine‐tuned LLMs. Conclusion Fine‐tuned LLMs demonstrated statistically significant improvements over original LLMs upon three clinically important tasks in radiation oncology. This study explored the feasibility of applying fine‐tuned LLMs in radiation oncology, inspiring further development of utilizing LLMs to assist with radiation oncology tasks.

Abstract Melanoma is characterized by its rapid progression and high mortality rates, making early and accurate detection essential for improving patient outcomes. This paper presents a comprehensive review of significant advancements in early melanoma detection, with a focus on integrating computer vision and deep learning techniques. This study investigates cutting-edge neural networks such as YOLO, GAN, Mask R-CNN, ResNet, and DenseNet to explore their application in enhancing early melanoma detection and diagnosis. These models were critically evaluated for their capacity to enhance dermatological imaging and diagnostic accuracy, crucial for effective melanoma treatment. Our research demonstrates that these AI technologies refine image analysis and feature extraction, and enhance processing capabilities in various clinical settings. Additionally, we emphasize the importance of comprehensive dermatological datasets such as PH2, ISIC, DERMQUEST, and MED-NODE, which are crucial for training and validating these sophisticated models. Integrating these datasets ensures that the AI systems are robust, versatile, and perform well under diverse conditions. The results of this study suggest that the integration of AI into melanoma detection marks a significant advancement in the field of medical diagnostics and is expected to have the potential to improve patient outcomes through more accurate and earlier detection methods. Future research should focus on enhancing these technologies further, integrating multimodal data, and improving AI decision interpretability to facilitate clinical adoption, thus transforming melanoma diagnostics into a more precise, personalized, and preventive healthcare service.

Abstract Genome-wide association studies (GWAS) have identified over 300 loci associated with the inflammatory bowel diseases (IBD), but putative causal genes for most are unknown. We conducted the largest disease-focused expression quantitative trait loci (eQTL) analysis using colon tissue from 252 IBD patients to determine genetic effects on gene expression and potential contribution to IBD. Combined with two non-IBD colon eQTL studies, we identified 194 potential target genes for 108 GWAS loci. eQTL in IBD tissue were enriched for IBD GWAS loci colocalizations, provided novel evidence for IBD-associated genes such as ABO and TNFRSF14, and identified additional target genes compared to non-IBD tissue eQTL. IBD-associated eQTL unique to diseased tissue had distinct regulatory and functional characteristics with increased effect sizes. Together, these highlight the importance of eQTL studies in diseased tissue for understanding functional consequences of genetic variants, and elucidating molecular mechanisms and regulation of key genes involved in IBD.

AbstractDNA damage plays a significant role in the tumorigenesis and progression of the disease. Abnormal DNA repair affects the therapy and prognosis of cancer. In this study, it is demonstrated that the deubiquitinase USP25 promotes non‐homologous end joining (NHEJ), which in turn contributes to chemoresistance in cancer. It is shown that USP25 deubiquitinates SHLD2 at the K64 site, which enhances its binding with REV7 and promotes NHEJ. Furthermore, USP25 deficiency impairs NHEJ‐mediated DNA repair and reduces class switch recombination (CSR) in USP25‐deficient mice. USP25 is overexpressed in a subset of colon cancers. Depletion of USP25 sensitizes colon cancer cells to IR, 5‐Fu, and cisplatin. TRIM25 is also identified, an E3 ligase, as the enzyme responsible for degrading USP25. Downregulation of TRIM25 leads to an increase in USP25 levels, which in turn induces chemoresistance in colon cancer cells. Finally, a peptide that disrupts the USP25‐SHLD2 interaction is successfully identified, impairing NHEJ and increasing sensitivity to chemotherapy in PDX model. Overall, these findings reveal USP25 as a critical effector of SHLD2 in regulating the NHEJ repair pathway and suggest its potential as a therapeutic target for cancer therapy.

Abstract Background Efficient and accurate delineation of organs at risk (OARs) is a critical procedure for treatment planning and dose evaluation. Deep learning‐based auto‐segmentation of OARs has shown promising results and is increasingly being used in radiation therapy. However, existing deep learning‐based auto‐segmentation approaches face two challenges in clinical practice: generalizability and human‐AI interaction. A generalizable and promptable auto‐segmentation model, which segments OARs of multiple disease sites simultaneously and supports on‐the‐fly human‐AI interaction, can significantly enhance the efficiency of radiation therapy treatment planning. Purpose Meta's segment anything model (SAM) was proposed as a generalizable and promptable model for next‐generation natural image segmentation. We further evaluated the performance of SAM in radiotherapy segmentation. Methods Computed tomography (CT) images of clinical cases from four disease sites at our institute were collected: prostate, lung, gastrointestinal, and head & neck. For each case, we selected the OARs important in radiotherapy treatment planning. We then compared both the Dice coefficients and Jaccard indices derived from three distinct methods: manual delineation (ground truth), automatic segmentation using SAM's ’segment anything’ mode, and automatic segmentation using SAM's ‘box prompt’ mode that implements manual interaction via live prompts during segmentation. Results Our results indicate that SAM's segment anything mode can achieve clinically acceptable segmentation results in most OARs with Dice scores higher than 0.7. SAM's box prompt mode further improves Dice scores by 0.1∼0.5. Similar results were observed for Jaccard indices. The results show that SAM performs better for prostate and lung, but worse for gastrointestinal and head & neck. When considering the size of organs and the distinctiveness of their boundaries, SAM shows better performance for large organs with distinct boundaries, such as lung and liver, and worse for smaller organs with less distinct boundaries, like parotid and cochlea. Conclusions Our results demonstrate SAM's robust generalizability with consistent accuracy in automatic segmentation for radiotherapy. Furthermore, the advanced box‐prompt method enables the users to augment auto‐segmentation interactively and dynamically, leading to patient‐specific auto‐segmentation in radiation therapy. SAM's generalizability across different disease sites and different modalities makes it feasible to develop a generic auto‐segmentation model in radiotherapy.

Background:Ibrutinib is approved in Europe and the US for the treatment of Waldenstrom macroglobulinemia (WM) with a median progression‐free survival (PFS) measured on several years. As PFS is commonly used for drug approval by regulatory entities, there is an unmet need of reliable surrogate markers for PFS that could help accelerating such process. This is of importance with novel BTK inhibitors currently in clinical trials.Aims:We hypothesized that the depth of response at 6 months can be predictive of PFS in patients with WM treated with the BTK inhibitor ibrutinib.Methods:We evaluated our hypothesis in two separate cohorts of WM patients treated with ibrutinib. The learning cohort was composed by participants of two prospective clinical trials at Dana‐Farber Cancer Institute (DFCI), Memorial Sloan Kettering Cancer Center and Stanford Cancer Center. The validation cohort was composed by consecutive WM patients treated with ibrutinib off clinical trial at DFCI and Mayo Clinic. Categorical responses at 6 months were assessed based on criteria from the 6thInternational Workshop for WM. PFS6 was defined as the time from the 6‐month response assessment until progression of disease or death from any cause. PFS6 curves were generated using the Kaplan‐Meier method and compared using the log‐rank test. Cox proportional hazard regression models were fitted for PFS6. Pvalues <0.05 were considered statistically significant.Results:The learning and validation cohorts were composed by 92 and 190 WM patients treated with ibrutinib, repsectively. There were no statistical differences between cohorts with regards to age, sex, serum IgM level, hemoglobin level, bone marrow involvement, prior treatment status, and MYD88 and CXCR4 mutational status (p>0.05 for all variables). The rates of partial response (PR) or better at 6 months in the learning and validation cohorts were 63% (54/86) and 72% (114/159), respectively (p = 0.15). The median PFS6 was not reached in the learning cohort and was 53 months (95% CI 44‐not reached) in the validation cohort. In the learning cohort, the median PFS6 for patients who attained a PR or better at 6 months was not reached, while for the patients who did not attained PR at 6 months was 48 months (95% CI 33‐not reached). The 3‐year PFS rate for patients who attained a PR or better at 6 months was 83% (95% CI 68–91%), while for patients who did not attained PR at 6 months was 58% (95% CI 33–76%) (Figure, left, log‐rank p = 0.02). Attaining PR or better at 6 months was associated with a hazard ratio (HR) or progressing or dying of 0.37 (95% CI 0.16–0.87; p = 0.02).In the validation cohort, the median PFS6 for patients who attained a PR or better at 6 months was 53 months (95% CI 44‐not reached), while for the patients who did not attained PR at 6 months was 32 months (95% CI 22‐not reached). The 3‐year PFS rate for patients who attained a PR or better at 6 months was 79% (95% CI 67–87%), while for patients who did not attained PR at 6 months was 50% (95% CI 26–70%) (Figure, right, log‐rank p = 0.04). Attaining PR or better at 6 months was associated with a hazard ratio (HR) or progressing or dying of 0.46 (95% CI 0.22–0.98; p = 0.04). Age, sex, serum IgM level, hemoglobin level, bone marrow involvement and prior treatment status were not prognostic of PFS6 in both cohorts.Summary/Conclusion:Attaining a PR or better at 6 months to ibrutinib is associated with longer median PFS and higher rates of 3‐year PFS in WM patients. Depth of response at 6 months could serve as a surrogate for PFS in clinical trials evaluating BTK inhibitors in WM patients.image

Abstract Microglia, the major resident immune cells in the central nervous system, serve as the frontline soldiers against cerebral ischemic injuries, possibly along with metabolic alterations. However, signaling pathways involved in the regulation of microglial immunometabolism in ischemic stroke remain to be further elucidated. In this study, using single‐nuclei RNA sequencing, a microglial subcluster up‐regulated in ischemic brain tissues is identified, with high expression of Igf1 and Trem2, neuroprotective transcriptional signature and enhanced oxidative phosphorylation. Microglial depletion by PLX3397 exacerbates ischemic brain damage, which is reversed by repopulating the microglia with high Igf1 and Trem2 phenotype. Mechanistically, Igf1 serves as one of the major down‐stream molecules of Trem2, and Trem2‐Igf1 signaling axis regulates microglial functional and metabolic profiles, exerting neuroprotective effects on ischemic stroke. Overexpression of Igf1 and supplementation of cyclocreatine restore microglial glucometabolic levels and cellular functions even in the absence of Trem2. These findings suggest that Trem2‐Igf1 signaling axis reprograms microglial immunometabolic profiles and shifts microglia toward a neuroprotective phenotype, which has promising therapeutic potential in treating ischemic stroke.

AbstractBackgroundAccurate and efficient dose calculation is essential for on‐line adaptive planning in proton therapy. Deep learning (DL) has shown promising dose prediction results in photon therapy. However, there is a scarcity of DL‐based dose prediction methods specifically designed for proton therapy. Successful dose prediction method for proton therapy should account for more challenging dose prediction problems in pencil beam scanning proton therapy (PBSPT) due to its sensitivity to heterogeneities.PurposeTo develop a DL‐based PBSPT dose prediction workflow with high accuracy and balanced complexity to support on‐line adaptive proton therapy clinical decision and subsequent replanning.MethodsPBSPT plans of 103 prostate cancer patients (93 for training and the other 10 for independent testing) and 83 lung cancer patients (73 for training and the other 10 for independent testing) previously treated at our institution were included in the study, each with computed tomography scans (CTs), structure sets, and plan doses calculated by the in‐house developed Monte‐Carlo dose engine (considered as the ground truth in the model training and testing). For the ablation study, we designed three experiments corresponding to the following three methods: (1) Experiment 1, the conventional region of interest (ROI) (composed of targets and organs‐at‐risk [OARs]) method. (2) Experiment 2, the beam mask (generated by raytracing of proton beams) method to improve proton dose prediction. (3) Experiment 3, the sliding window method for the model to focus on local details to further improve proton dose prediction. A fully connected 3D‐Unet was adopted as the backbone. Dose volume histogram (DVH) indices, 3D Gamma passing rates with a criterion of 3%/3 mm/10%, and dice coefficients for the structures enclosed by the iso‐dose lines between the predicted and the ground truth doses were used as the evaluation metrics. The calculation time for each proton dose prediction was recorded to evaluate the method's efficiency.ResultsCompared to the conventional ROI method, the beam mask method improved the agreement of DVH indices for both targets and OARs and the sliding window method further improved the agreement of the DVH indices (for lung cancer, CTV D98 absolute deviation: 0.74 ± 0.18 vs. 0.57 ± 0.21 vs. 0.54 ± 0.15 Gy[RBE], ROI vs. beam mask vs. sliding window methods, respectively). For the 3D Gamma passing rates in the target, OARs, and BODY (outside target and OARs), the beam mask method improved the passing rates in these regions and the sliding window method further improved them (for prostate cancer, targets: 96.93% ± 0.53% vs. 98.88% ± 0.49% vs. 99.97% ± 0.07%, BODY: 86.88% ± 0.74% vs. 93.21% ± 0.56% vs. 95.17% ± 0.59%). A similar trend was also observed for the dice coefficients. This trend was especially remarkable for relatively low prescription isodose lines (for lung cancer, 10% isodose line dice: 0.871 ± 0.027 vs. 0.911 ± 0.023 vs. 0.927 ± 0.017). The dose predictions for all the testing cases were completed within 0.25 s.ConclusionsAn accurate and efficient deep learning‐augmented proton dose prediction framework has been developed for PBSPT, which can predict accurate dose distributions not only inside but also outside ROI efficiently. The framework can potentially further reduce the initial planning and adaptive replanning workload in PBSPT.

Across multiancestry groups, we analyzed Human Leukocyte Antigen (HLA) associations in over 176,000 individuals with Parkinson’s disease (PD) and Alzheimer’s disease (AD) versus controls. We demonstrate that the two diseases share the same protective association at the HLA locus. HLA-specific fine-mapping showed that hierarchical protective effects of HLA-DRB1 *04 subtypes best accounted for the association, strongest with HLA-DRB1 *04:04 and HLA-DRB1 *04:07, and intermediary with HLA-DRB1 *04:01 and HLA-DRB1 *04:03. The same signal was associated with decreased neurofibrillary tangles in postmortem brains and was associated with reduced tau levels in cerebrospinal fluid and to a lower extent with increased Aβ42. Protective HLA-DRB1 *04 subtypes strongly bound the aggregation-prone tau PHF6 sequence, however only when acetylated at a lysine (K311), a common posttranslational modification central to tau aggregation. An HLA-DRB1 *04-mediated adaptive immune response decreases PD and AD risks, potentially by acting against tau, offering the possibility of therapeutic avenues.