Exploring combinations of dimensionality reduction, transfer learning, and regularization methods for predicting binary phenotypes with transcriptomic data

BACKGROUND

Numerous transcriptomic-based models have been developed to predict or understand the fundamental mechanisms driving biological phenotypes. However, few models have successfully transitioned into clinical practice due to challenges associated with generalizability and interpretability. To address these issues, researchers have turned to dimensionality reduction methods and have begun implementing transfer learning approaches.

METHODS

In this study, we aimed to determine the optimal combination of dimensionality reduction and regularization methods for predictive modeling. We applied seven dimensionality reduction methods to various datasets, including two supervised methods (linear optimal low-rank projection and low-rank canonical correlation analysis), two unsupervised methods [principal component analysis and consensus independent component analysis (c-ICA)], and three methods [autoencoder (AE), adversarial variational autoencoder, and c-ICA] within a transfer learning framework, trained on > 140,000 transcriptomic profiles. To assess the performance of the different combinations, we used a cross-validation setup encapsulated within a permutation testing framework, analyzing 30 different transcriptomic datasets with binary phenotypes. Furthermore, we included datasets with small sample sizes and phenotypes of varying degrees of predictability, and we employed independent datasets for validation.

RESULTS

Our findings revealed that regularized models without dimensionality reduction achieved the highest predictive performance, challenging the necessity of dimensionality reduction when the primary goal is to achieve optimal predictive performance. However, models using AE and c-ICA with transfer learning for dimensionality reduction showed comparable performance, with enhanced interpretability and robustness of predictors, compared to models using non-dimensionality-reduced data.

CONCLUSION

These findings offer valuable insights into the optimal combination of strategies for enhancing the predictive performance, interpretability, and generalizability of transcriptomic-based models.

Genomic Engineering of Oral Keratinocytes to Establish In Vitro Oral Potentially Malignant Disease Models as a Platform for Treatment Investigation

Precancerous cells in the oral cavity may appear as oral potentially malignant disorders, but they may also present as dysplasia without visual manifestation in tumor-adjacent tissue. As it is currently not possible to prevent the malignant transformation of these oral precancers, new treatments are urgently awaited. Here, we generated precancer culture models using a previously established method for the generation of oral keratinocyte cultures and incorporated CRISPR/Cas9 editing. The generated cell lines were used to investigate the efficacy of a set of small molecule inhibitors. Tumor-adjacent mucosa and oral leukoplakia biopsies were cultured and genetically characterized. Mutations were introduced in CDKN2A and TP53 using CRISPR/Cas9 and combined with the ectopic activation of telomerase to generate cell lines with prolonged proliferation. The method was tested in normal oral keratinocytes and tumor-adjacent biopsies and subsequently applied to a large set of oral leukoplakia biopsies. Finally, a subset of the immortalized cell lines was used to assess the efficacy of a set of small molecule inhibitors. Culturing and genomic engineering was highly efficient for normal and tumor-adjacent oral keratinocytes, but success rates in oral leukoplakia were remarkably low. Knock-out of CDKN2A in combination with either the activation of telomerase or knock-out of TP53 seemed a prerequisite for immortalization. Prolonged culturing was accompanied by additional genetic aberrations in these cultures. The generated cell lines were more sensitive than normal keratinocytes to small molecule inhibitors of previously identified targets. In conclusion, while very effective for normal keratinocytes and tumor-adjacent biopsies, the success rate of oral leukoplakia cell culturing methods was very low. Genomic engineering enabled the prolonged culturing of OL-derived keratinocytes but was associated with acquired genetic changes. Further studies are required to assess to what extent the immortalized cultures faithfully represent characteristics of the cells in vivo.

Origin and outcome of metastatic tumours to the testes: a nationwide study

OBJECTIVES

To perform a retrospective cohort analysis for metastatic tumours in the testes to explore the timing, presentation and prognosis of this particular type of metastases and the factors that influence outcome.

PATIENTS AND METHODS

A nationwide retrospective review of pathology reports of patients with pathologically confirmed metastases to the testis between 1991 and 2021 was performed. Data were collected from the Dutch nationwide pathology databank (PALGA) and the Netherlands Cancer Registry. Log-rank testing and Kaplan-Meier analyses were used to assess overall survival (OS), and Cox proportional hazard models were used for multivariate survival analysis.

RESULTS

A total of 175 patients with a testicular metastasis were included. The median (range) age at diagnosis of testicular metastasis was 67 (3-88) years. Testicular metastases originated from a variety of primary tumours, although most frequently from the prostate (40.6%), kidney (13.7%), colon (10.3%), bladder (7.4%) and skin (5.7%). Synchronous testicular metastasis was detected in 53 cases, while 114 metachronous lesions were found after a median (interquartile range) interval of 22 (1-53) months after the original cancer diagnosis. OS after the diagnosis of a testicular metastasis was poor, with a median survival of 14.2 months (95% confidence interval 10.2-18.3). Primary tumour origin was an independent factor for survival, with worst survival for patients with primary skin, bladder and colon cancer.

CONCLUSION

Testicular metastases are very uncommon and arise mainly from primary tumours anatomically close to the testes. Most patients develop metachronous testicular metastasis at an oligometastatic disease stage. These metastases are invariably associated with poor survival.

NnU-Net versus mesh growing algorithm as a tool for the robust and timely segmentation of neurosurgical 3D images in contrast-enhanced T1 MRI scans

PURPOSE

This study evaluates the nnU-Net for segmenting brain, skin, tumors, and ventricles in contrast-enhanced T1 (T1CE) images, benchmarking it against an established mesh growing algorithm (MGA).

METHODS

We used 67 retrospectively collected annotated single-center T1CE brain scans for training models for brain, skin, tumor, and ventricle segmentation. An additional 32 scans from two centers were used test performance compared to that of the MGA. The performance was measured using the Dice-Sørensen coefficient (DSC), intersection over union (IoU), 95th percentile Hausdorff distance (HD95), and average symmetric surface distance (ASSD) metrics, with time to segment also compared.

RESULTS

The nnU-Net models significantly outperformed the MGA (p < 0.0125) with a median brain segmentation DSC of 0.971 [95CI: 0.945-0.979], skin: 0.997 [95CI: 0.984-0.999], tumor: 0.926 [95CI: 0.508-0.968], and ventricles: 0.910 [95CI: 0.812-0.968]. Compared to the MGA's median DSC for brain: 0.936 [95CI: 0.890, 0.958], skin: 0.991 [95CI: 0.964, 0.996], tumor: 0.723 [95CI: 0.000-0.926], and ventricles: 0.856 [95CI: 0.216-0.916]. NnU-Net performance between centers did not significantly differ except for the skin segmentations Additionally, the nnU-Net models were faster (mean: 1139 s [95CI: 685.0-1616]) than the MGA (mean: 2851 s [95CI: 1482-6246]).

CONCLUSIONS

The nnU-Net is a fast, reliable tool for creating automatic deep learning-based segmentation pipelines, reducing the need for extensive manual tuning and iteration. The models are able to achieve this performance despite a modestly sized training set. The ability to create high-quality segmentations in a short timespan can prove invaluable in neurosurgical settings.

Detection of dendritic cell subsets in the tumor microenvironment by multiplex immunohistochemistry

Dendritic cells (DCs) are essential in antitumor immunity. In humans, three main DC subsets are defined: two types of conventional DCs (cDC1s and cDC2s) and plasmacytoid DCs (pDCs). To study DC subsets in the tumor microenvironment (TME), it is important to correctly identify them in tumor tissues. Tumor-derived DCs are often analyzed in cell suspensions in which spatial information about DCs which can be important to determine their function within the TME is lost. Therefore, we developed the first standardized and optimized multiplex immunohistochemistry panel, simultaneously detecting cDC1s, cDC2s, and pDCs within their tissue context. We report on this panel's development, validation, and quantitative analysis. A multiplex immunohistochemistry panel consisting of CD1c, CD303, X-C motif chemokine receptor 1, CD14, CD19, a tumor marker, and DAPI was established. The ImmuNet machine learning pipeline was trained for the detection of DC subsets. The performance of ImmuNet was compared with conventional cell phenotyping software. Ultimately, frequencies of DC subsets within several tumors were defined. In conclusion, this panel provides a method to study cDC1s, cDC2s, and pDCs in the spatial context of the TME, which supports unraveling their specific roles in antitumor immunity.

Evaluation Metrics for Augmented Reality in Neurosurgical Preoperative Planning, Surgical Navigation, and Surgical Treatment Guidance: A Systematic Review

BACKGROUND AND OBJECTIVE

Recent years have shown an advancement in the development of augmented reality (AR) technologies for preoperative visualization, surgical navigation, and intraoperative guidance for neurosurgery. However, proving added value for AR in clinical practice is challenging, partly because of a lack of standardized evaluation metrics. We performed a systematic review to provide an overview of the reported evaluation metrics for AR technologies in neurosurgical practice and to establish a foundation for assessment and comparison of such technologies.

METHODS

PubMed, Embase, and Cochrane were searched systematically for publications on assessment of AR for cranial neurosurgery on September 22, 2022. The findings were reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines.

RESULTS

The systematic search yielded 830 publications; 114 were screened full text, and 80 were included for analysis. Among the included studies, 5% dealt with preoperative visualization using AR, with user perception as the most frequently reported metric. The majority (75%) researched AR technology for surgical navigation, with registration accuracy, clinical outcome, and time measurements as the most frequently reported metrics. In addition, 20% studied the use of AR for intraoperative guidance, with registration accuracy, task outcome, and user perception as the most frequently reported metrics.

CONCLUSION

For quality benchmarking of AR technologies in neurosurgery, evaluation metrics should be specific to the risk profile and clinical objectives of the technology. A key focus should be on using validated questionnaires to assess user perception; ensuring clear and unambiguous reporting of registration accuracy, precision, robustness, and system stability; and accurately measuring task performance in clinical studies. We provided an overview suggesting which evaluation metrics to use per AR application and innovation phase, aiming to improve the assessment of added value of AR for neurosurgical practice and to facilitate the integration in the clinical workflow.

Multi-class glioma segmentation on real-world data with missing MRI sequences: comparison of three deep learning algorithms

This study tests the generalisability of three Brain Tumor Segmentation (BraTS) challenge models using a multi-center dataset of varying image quality and incomplete MRI datasets. In this retrospective study, DeepMedic, no-new-Unet (nn-Unet), and NVIDIA-net (nv-Net) were trained and tested using manual segmentations from preoperative MRI of glioblastoma (GBM) and low-grade gliomas (LGG) from the BraTS 2021 dataset (1251 in total), in addition to 275 GBM and 205 LGG acquired clinically across 12 hospitals worldwide. Data was split into 80% training, 5% validation, and 15% internal test data. An additional external test-set of 158 GBM and 69 LGG was used to assess generalisability to other hospitals' data. All models' median Dice similarity coefficient (DSC) for both test sets were within, or higher than, previously reported human inter-rater agreement (range of 0.74-0.85). For both test sets, nn-Unet achieved the highest DSC (internal = 0.86, external = 0.93) and the lowest Hausdorff distances (10.07, 13.87 mm, respectively) for all tumor classes (p < 0.001). By applying Sparsified training, missing MRI sequences did not statistically affect the performance. nn-Unet achieves accurate segmentations in clinical settings even in the presence of incomplete MRI datasets. This facilitates future clinical adoption of automated glioma segmentation, which could help inform treatment planning and glioma monitoring.

The clinical importance of the host anti-tumour reaction patterns in regional tumour draining lymph nodes in patients with locally advanced resectable gastric cancer: a systematic review and meta-analysis

BACKGROUND

The status of regional tumour draining lymph nodes (LN) is crucial for prognostic evaluation in gastric cancer (GaC) patients. Changes in lymph node microarchitecture, such as follicular hyperplasia (FH), sinus histiocytosis (SH), or paracortical hyperplasia (PH), may be triggered by the anti-tumour immune response. However, the prognostic value of these changes in GaC patients is unclear.

METHODS

A systematic search in multiple databases was conducted to identify studies on the prognostic value of microarchitecture changes in regional tumour-negative and tumour-positive LNs measured on histopathological slides. Since the number of GaC publications was very limited, the search was subsequently expanded to include junctional and oesophageal cancer (OeC).

RESULTS

A total of 28 articles (17 gastric cancer, 11 oesophageal cancer) met the inclusion criteria, analyzing 26,503 lymph nodes from 3711 GaC and 1912 OeC patients. The studies described eight different types of lymph node microarchitecture changes, categorized into three patterns: hyperplasia (SH, FH, PH), cell-specific infiltration (dendritic cells, T cells, neutrophils, macrophages), and differential gene expression. Meta-analysis of five GaC studies showed a positive association between SH in tumour-negative lymph nodes and better 5-year overall survival. Pooled risk ratios for all LNs showed increased 5-year overall survival for the presence of SH and PH.

CONCLUSIONS

This systematic review suggests that sinus histiocytosis and paracortical hyperplasia in regional tumour-negative lymph nodes may provide additional prognostic information for gastric and oesophageal cancer patients. Further studies are needed to better understand the lymph node reaction patterns and explore their impact of chemotherapy treatment and immunotherapy efficacy.

The role of differentiated dysplasia in the prediction of malignant transformation of oral leukoplakia

OBJECTIVE

Oral leukoplakia is the most common oral potentially malignant disorder. Malignant transformation of oral leukoplakia occurs at an annual rate of 1%-7%. WHO-defined classic epithelial dysplasia is an important predictor of malignant transformation of oral leukoplakia, but we have previously shown in a proof of concept study that prediction improves by incorporation of an architectural pattern of dysplasia, also coined as differentiated dysplasia. We aimed to analyze this finding in a larger cohort of patients.

METHOD

For this retrospective study 176 oral leukoplakia patients were included. Biopsies for all patients were assessed for the presence of dysplasia and analyzed for cytokeratin 13 and 17 expression. Moreover, the inter-observer agreement for the diagnosis of differentiated dysplasia was determined.

RESULTS

In total, 33 of 176 patients developed oral squamous cell carcinoma during follow-up. Presence of classic epithelial dysplasia increased cancer risk two-fold (HR = 2.18, p = 0.026). Lesions without classic epithelial dysplasia could be further risk-stratified by the presence of differentiated dysplasia (HR = 7.36, p < 0.001). Combined classic epithelial and differentiated dysplasia imparted a seven-fold increased risk of malignant transformation (7.34, p = 0.001). Inter-observer agreement for the diagnosis of dysplasia, including differentiated dysplasia, was moderate (κ = 0.56, p < 0.001).

DISCUSSION

This study emphasizes the importance of the recognition of the architectural pattern of differentiated dysplasia as a separate entity for risk prediction of malignant transformation of oral leukoplakia. Presence of any pattern of dysplasia results in accurate prediction of malignant transformation risk of oral leukoplakia.

Sensitivity of an AI method for [18F]FDG PET/CT outcome prediction of diffuse large B-cell lymphoma patients to image reconstruction protocols

BACKGROUND

Convolutional neural networks (CNNs), applied to baseline [18F]-FDG PET/CT maximum intensity projections (MIPs), show potential for treatment outcome prediction in diffuse large B-cell lymphoma (DLBCL). The aim of this study is to investigate the robustness of CNN predictions to different image reconstruction protocols. Baseline [18F]FDG PET/CT scans were collected from 20 DLBCL patients. EARL1, EARL2 and high-resolution (HR) protocols were applied per scan, generating three images with different image qualities. Image-based transformation was applied by blurring EARL2 and HR images to generate EARL1 compliant images using a Gaussian filter of 5 and 7 mm, respectively. MIPs were generated for each of the reconstructions, before and after image transformation. An in-house developed CNN predicted the probability of tumor progression within 2 years for each MIP. The difference in probabilities per patient was then calculated between both EARL2 and HR with respect to EARL1 (delta probabilities or ΔP). We compared these to the probabilities obtained after aligning the data with ComBat using the difference in median and interquartile range (IQR).

RESULTS

CNN probabilities were found to be sensitive to different reconstruction protocols (EARL2 ΔP: median = 0.09, interquartile range (IQR) = [0.06, 0.10] and HR ΔP: median = 0.1, IQR = [0.08, 0.16]). Moreover, higher resolution images (EARL2 and HR) led to higher probability values. After image-based and ComBat transformation, an improved agreement of CNN probabilities among reconstructions was found for all patients. This agreement was slightly better after image-based transformation (transformed EARL2 ΔP: median = 0.022, IQR = [0.01, 0.02] and transformed HR ΔP: median = 0.029, IQR = [0.01, 0.03]).

CONCLUSION

Our CNN-based outcome predictions are affected by the applied reconstruction protocols, yet in a predictable manner. Image-based harmonization is a suitable approach to harmonize CNN predictions across image reconstruction protocols.

An artificial intelligence method using FDG PET to predict treatment outcome in diffuse large B cell lymphoma patients

Convolutional neural networks (CNNs) may improve response prediction in diffuse large B-cell lymphoma (DLBCL). The aim of this study was to investigate the feasibility of a CNN using maximum intensity projection (MIP) images from 18F-fluorodeoxyglucose (18F-FDG) positron emission tomography (PET) baseline scans to predict the probability of time-to-progression (TTP) within 2 years and compare it with the International Prognostic Index (IPI), i.e. a clinically used score. 296 DLBCL 18F-FDG PET/CT baseline scans collected from a prospective clinical trial (HOVON-84) were analysed. Cross-validation was performed using coronal and sagittal MIPs. An external dataset (340 DLBCL patients) was used to validate the model. Association between the probabilities, metabolic tumour volume and Dmaxbulk was assessed. Probabilities for PET scans with synthetically removed tumors were also assessed. The CNN provided a 2-year TTP prediction with an area under the curve (AUC) of 0.74, outperforming the IPI-based model (AUC = 0.68). Furthermore, high probabilities (> 0.6) of the original MIPs were considerably decreased after removing the tumours (< 0.4, generally). These findings suggest that MIP-based CNNs are able to predict treatment outcome in DLBCL.