Automated Multi-Objective ER-rule ensemble model for Locoregional Recurrence Prediction in Head and Neck Cancer.

Head and neck (HN) cancer is an endemic disease in Taiwan, China. Locally recurrent HN cancer after full-dose irradiation poses a therapeutic challenge, and boron neutron capture therapy (BNCT) may be a solution that could provide durable local control with tolerable toxicity. The Tsing-Hua Open Pool Reactor (THOR) at National Tsing-Hua University in Hsin-Chu, provides a high-quality epithermal neutron source for basic and clinical BNCT research. Our first clinical trial, entitled “A phase I/II trial of boron neutron capture therapy for recurrent head and neck cancer at THOR”, was carried out between 2010 and 2013. A total of 17 patients with 23 recurrent HN tumors who had received high-dose photon irradiation were enrolled in the study. The fructose complex of l-boronophenylalanine was used as a boron carrier, and a two-fraction BNCT treatment regimen at 28-day intervals was used for each patient. Toxicity was acceptable, and although the response rate was high (12/17), re-recurrence within or near the radiation site was common. To obtain better local control, another clinical trial entitled “A phase I/II trial of boron neutron capture therapy combined with image-guided intensity-modulated radiotherapy (IG-IMRT) for locally recurrent HN cancer” was initiated in 2014. The first administration of BNCT was performed according to our previous protocol, and IG-IMRT was initiated 28 days after BNCT. As of May 2017, seven patients have been treated with this combination. The treatment-related toxicity was similar to that previously observed with two BNCT applications. Three patients had a complete response, but locoregional recurrence was the major cause of failure despite initially good responses. Future clinical trials combining BNCT with other local or systemic treatments will be carried out for recurrent HN cancer patients at THOR.

Comparison of Machine Learning and Deep Learning Methods for the Prediction of Osteoradionecrosis Resulting from Head and Neck Cancer Radiation Therapy

Machine Learning Methods Uncover Radiomorphologic Dose Patterns in Salivary Glands that Predict Xerostomia in Patients with Head and Neck Cancer

Objective. Deep learning shows promise in autosegmentation of head and neck cancer (HNC) primary tumours (GTV-T) and nodal metastases (GTV-N). However, errors such as including non-tumour regions or missing nodal metastases still occur. Conventional methods often make overconfident predictions, compromising reliability. Incorporating uncertainty estimation, which provides calibrated confidence intervals can address this issue. Our aim was to investigate the efficacy of various uncertainty estimation methods in improving segmentation reliability. We evaluated their confidence levels in voxel predictions and ability to reveal potential segmentation errors. Approach. We retrospectively collected data from 567 HNC patients with diverse cancer sites and multi-modality images (CT, PET, T1-, and T2-weighted MRI) along with their clinical GTV-T/N delineations. Using the nnUNet 3D segmentation pipeline, we compared seven uncertainty estimation methods, evaluating them based on segmentation accuracy (Dice similarity coefficient, DSC), confidence calibration (Expected Calibration Error, ECE), and their ability to reveal segmentation errors (Uncertainty-Error overlap using DSC, UE-DSC). Main results. Evaluated on the hold-out test dataset (n = 97), the median DSC scores for GTV-T and GTV-N segmentation across all uncertainty estimation methods had a narrow range, from 0.73 to 0.76 and 0.78 to 0.80, respectively. In contrast, the median ECE exhibited a wider range, from 0.30 to 0.12 for GTV-T and 0.25 to 0.09 for GTV-N. Similarly, the median UE-DSC also ranged broadly, from 0.21 to 0.38 for GTV-T and 0.22 to 0.36 for GTV-N. A probabilistic network—PhiSeg method consistently demonstrated the best performance in terms of ECE and UE-DSC. Significance. Our study highlights the importance of uncertainty estimation in enhancing the reliability of deep learning for autosegmentation of HNC GTV. The results show that while segmentation accuracy can be similar across methods, their reliability, measured by calibration error and uncertainty-error overlap, varies significantly. Used with visualisation maps, these methods may effectively pinpoint uncertainties and potential errors at the voxel level.

Head and Neck Cancer Genes Predictive of Radioresistance and Detriment to Local Control

This study presents a retrospective analysis (efficacy and toxicity) of outcomes in patients with unresectable recurrence of previously irradiated head and neck (H&N) cancers treated with proton therapy. Locoregional recurrence is the main pattern of failure in the treatment of H&N cancers. Proton re-irradiation in patients with relapse after prior radiotherapy might be valid as promising as a challenging treatment option. From November 2015 to January 2020, 30 patients with in-field recurrence of head and neck cancer, who were not suitable for surgery due to medical contraindications, tumor localization, or extent, received re-irradiation with intensity-modulated proton therapy (IMPT). Sites of retreatment included the aerodigestive tract (60%) and the base of skull (40%). The median total dose of prior radiotherapy was 55.0 Gy. The median time to the second course was 38 months. The median re-irradiated tumor volume was 158.1 cm3. Patients were treated with 2.0, 2.4, and 3.0 GyRBE per fraction, with a median equivalent dose (EQD2) of 57.6 Gy (α/β = 10). Radiation-induced toxicity was recorded according to the RTOG/EORTC criteria. The 1- and 2-year local control (LC), progression-free survival (PFS), and overall survival (OS) were 52.6/21.0, 21.9/10.9, and 73.4/8.4%, respectively, with a median follow-up time of 21 months. The median overall survival was 16 months. Acute grade 3 toxicity was observed in one patient (3.3%). There were five late severe side effects (16.6%), with one death associated with re-irradiation. Re-irradiation with a proton beam can be considered a safe and efficient treatment even for a group of patients with unresectable recurrent H&N cancers.

Repeat re-irradiation with interstitial HDR-brachytherapy for an in-field isolated nodal recurrence in a patient with HPV-positive squamous cell carcinoma of the head and neck

Cancer types are typically categorized according to the cell of origin, but within these groupings there exists vast heterogeneity. Thus, subtypes based on molecular features are often defined that have clinical utility as prognostic biomarkers, aid in selection of therapeutic strategies, or are associated with treatment response. Head and neck cancer (HNC) is the seventh most common cancer globally and projected to have 54,540 new cases in the U.S. in 2023. Within the US, HPV(+) HNC is now more prevalent than HPV(+) cervical cancer, and it is expected to continue to rise. Although there is wide morphologic and epigenetic diversity within HPV(+) HNC, tumor subtyping is not yet widely used for this cancer population, largely because of HPV(+) HNC’s unique tumorigenesis process. Unlike other cancers that are driven by specific genetic signatures, HNC is driven by viral gene mechanisms. Given the large degree of heterogeneity among HPV(+) HNC cases and the lack of key mutations to define subtypes, the classification of HPV(+) HNSC subtypes requires a more sophisticated approach. In this study, we introduce a robust, ensemble machine learning (ML) classifier for subtyping HPV(+) HNC that was trained on multiple cohorts and gene feature sets from RNA-seq data, and rigorously tested to ensure high reproducibility. The results classify HPV(+) HNC into the two main recognized subtypes: IMU (immune strong) and KRT(highly keratinized). Using a cohort of 227 patients, we show that the IMU/KRT classification is highly correlated with key clinically relevant variables in HNC. Of 41 molecular, clinical, and epidemiologic variables tested, 24 significantly associated with subtype. The IMU subtype was significantly associated with CD8 T-cells (p-value = 4.21 × 10−9), Dendritic cells (p-value = 1.65 × 10−8), B-cells (p-value = 2.20E−8), and epithelial to mesenchymal transition (p-value = 0.00013146). The KRT subtype is significantly associated with keratinization (p-value = 7.11 × 10−9), HPV integration (p-value = 3.14 × 10−6), radiation resistance (p-value = 0.00307), female sex (p-value = 0.00641), and high T stage (p-value = 0.0324). Genetic, epigenetic, and HPV gene-related variables were also among those significantly associated with subtype. HPV(+) HNC subtypes have been shown to be associated with survival, with KRT-like patients having worse clinical outcomes than IMU. Given the different carcinogenic processes underlying IMU and KRT tumors, our ensemble ML subtype classifier web tool will help inform future studies of HPV(+) HNC. Future work to assess how HPV(+) subtypes can be incorporated into precision treatment strategies is well-motivated by our findings. Citation Format: Bailey F. Garb, Shiting Li, Tingting Qin, Elizabeth Lopez, Sarah Soppe, Snehal Patil, Laura Rozek, Nisha D'Silva, Maureen Sartor. Tumor subtype classification of HPV-associated head and neck cancers is central to key clinically relevant variables [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 3491.

Simple SummaryIn head and neck cancer surgery, blood transfusion is required occasionally due to patients’ underlying conditions and perioperative blood loss during surgical resection. However, transfusion is associated with immunosuppression, also known as the term “transfusion-related immune modulation (TRIM)”, which could lead to worse cancer prognoses. The purpose of the study is to assess the association between perioperative blood transfusion and head and neck cancer recurrence and mortality. Our findings showed that blood transfusion was significantly associated with both cancer recurrence and mortality after head and neck cancer surgery.Background: The association between perioperative blood transfusion and cancer prognosis in patients with head and neck cancer (HNC) receiving surgery remains controversial. Methods: We designed a retrospective observational study of patients with HNC undergoing tumor resection surgery from 2014 to 2017 and followed them up until June 2020. An inverse probability of treatment weighting (IPTW) was applied to balance baseline patient characteristics in the exposed and unexposed groups. COX regression was used for the evaluation of tumor recurrence and overall survival. Results: A total of 683 patients were included; 192 of them (28.1%) received perioperative packed RBC transfusion. Perioperative blood transfusion was significantly associated with HNC recurrence (IPTW adjusted HR: 1.37, 95% CI: 1.1–1.7, p = 0.006) and all-cause mortality (IPTW adjusted HR: 1.37, 95% CI: 1.07–1.74, p = 0.011). Otherwise, there was an increased association with cancer recurrence in a dose-dependent manner. Conclusion: Perioperative transfusion was associated with cancer recurrence and mortality after HNC tumor surgery.

The locoregional recurrence rate after treatment of head and neck cancer (HNC) is known to be about 40%, and recurrence of cancer is the major factor directly related to the survival of patients. Recurrent HNC has different biological characteristics and tumor microenvironment from those of index cancer. And it subsequently exhibits pro-tumoral and treatment-resistant traits, which leads to difficulties in selecting salvage treatments and followed by dismal prognosis. Furthermore, since which salvage treatment can be selected and what the result of it will be determined by the prior treatment, there should be careful consideration in the initial therapeutic strategy. In this review, currently used treatment methods and results for locoregionally recurrent HNC are summarized, and considerations for each treatment based on the clinical and biomolecular characteristics of recurrent HNC are discussed. In addition, this review contains introductions of new therapeutic strategies including recent clinical trials and a perspective on the future direction for treatment of locoregionally recurrent HNC.

Aim: To assess the efficacy of reirradiation in locoregionally recurrent head and neck cancer (HNC) and to describe results in our center in relation to other published data among similar group of patients. Methods: The medical records of 28 patients with HNC who received reirradiation with or without chemotherapy for loco-regional recurrence between 2005 and 2013 were reviewed. They were evaluated for; toxicity profile, overall survival (OS) and progression free survival (PFS). Results: The median reirradiation dose was 50 Gy (range 40-60 Gy) and median radiation cumulative dose was 119 (range 113 -120). An overall response rate was seen in 36% of patients with only 3 patients showed complete response. The median OS was 9 months with 1-and 2-year survival rates being 34.1% and 10.6%. The OS and PFS were significantly better in patients who were treated with chemotherapy concomitant with radiation and received higher radiation dose. Grade 3 mucositis and skin reactions were seen in 24 % and 14% of patients, respectively. Conclusion: Reirradiation appears to be feasible in patients with recurrent HNC treated previously with radiation. The benefit of concurrent chemotherapy with reirradiation is expected. Our results are subject to limitations from the retrospective nature of the analysis, the relatively small number, and improper selection of patients.

BackgroundThe accurate prediction of genomic breeding values is central to genomic selection in both plant and animal breeding studies. Genomic prediction involves the use of thousands of molecular markers spanning the entire genome and therefore requires methods able to efficiently handle high dimensional data. Not surprisingly, machine learning methods are becoming widely advocated for and used in genomic prediction studies. These methods encompass different groups of supervised and unsupervised learning methods. Although several studies have compared the predictive performances of individual methods, studies comparing the predictive performance of different groups of methods are rare. However, such studies are crucial for identifying (i) groups of methods with superior genomic predictive performance and assessing (ii) the merits and demerits of such groups of methods relative to each other and to the established classical methods. Here, we comparatively evaluate the genomic predictive performance and informally assess the computational cost of several groups of supervised machine learning methods, specifically, regularized regression methods, deep, ensemble and instance-based learning algorithms, using one simulated animal breeding dataset and three empirical maize breeding datasets obtained from a commercial breeding program.ResultsOur results show that the relative predictive performance and computational expense of the groups of machine learning methods depend upon both the data and target traits and that for classical regularized methods, increasing model complexity can incur huge computational costs but does not necessarily always improve predictive accuracy. Thus, despite their greater complexity and computational burden, neither the adaptive nor the group regularized methods clearly improved upon the results of their simple regularized counterparts. This rules out selection of one procedure among machine learning methods for routine use in genomic prediction. The results also show that, because of their competitive predictive performance, computational efficiency, simplicity and therefore relatively few tuning parameters, the classical linear mixed model and regularized regression methods are likely to remain strong contenders for genomic prediction.ConclusionsThe dependence of predictive performance and computational burden on target datasets and traits call for increasing investments in enhancing the computational efficiency of machine learning algorithms and computing resources.

This scoping review aimed to provide both researchers and practitioners with an overview of how machine learning (ML) methods are applied to infrared spectroscopy for the diagnosis and prognosis of head and neck precancer and cancer. A subject headings and keywords search was conducted in MEDLINE, Embase, and Scopus on 14 January 2024, using predefined search algorithms targeting studies that integrated infrared spectroscopy and ML methods in head and neck precancer/cancer research. The results were managed through the COVIDENCE systematic review platform. Fourteen studies met the eligibility criteria, which were defined by IR spectroscopy techniques, ML methodology, and a focus on head and neck precancer/cancer research involving human subjects. The IR spectroscopy techniques used in these studies included Fourier transform infrared (FTIR) spectroscopy and imaging, attenuated total reflection-FTIR, near-infrared spectroscopy, and synchrotron-based infrared microspectroscopy. The investigated human biospecimens included tissues, exfoliated cells, saliva, plasma, and urine samples. ML methods applied in the studies included linear discriminant analysis (LDA), principal component analysis with LDA, partial least squares discriminant analysis, orthogonal partial least squares discriminant analysis, support vector machine, extreme gradient boosting, canonical variate analysis, and deep reinforcement neural network. For oral cancer diagnosis applications, the highest sensitivity and specificity were reported to be 100%, the highest accuracy was reported to be 95-96%, and the highest area under the curve score was reported to be 0.99. For oral precancer prognosis applications, the highest sensitivity and specificity were reported to be 84% and 79%, respectively. This review highlights the promising potential of integrating infrared spectroscopy with ML methods for diagnosing and prognosticating head and neck precancer and cancer. However, the limited sample sizes in existing studies restrict generalizability of the study findings. Future research should prioritize larger datasets and the development of advanced ML models to enhance reliability and robustness of these tools.

Convolutional Neural Network Learning from Combinations of RT Dose Distribution, CT and PET Improves Predicting Locoregional Recurrence for Head and Neck Cancer

Ensemble deep learning in speech signal tasks: A review