Marc Eric Miquel

Accurate medical image segmentation plays a vital role in assisting diagnosis with quantifiable visual evidence. Due to the complex structure and diverse patterns in medical images, it is crucial to capture both short and long-range pixel relations. While transformers are adept at modeling long-range spatial dependencies in images, they struggle with learning local pixel relationships. To address this, we propose a deep learning network named GoLoCo-Net incorporating a dual decoder structure. More specifically, one decoder entails a Contextual Attention Feature Enhancement (CAFE) module to enhance the features for a broader capture of local and global contexts, whereas the other uses a Global-Guide-Local Feature (GGLF) module that leverages high-level features to enrich low-level features with a global context. The proposed method is evaluated on two dynamic MRI datasets and one multi-organ CT dataset. Experimental results show that the model achieves state-of-the-art performance across all three datasets. The code is available: https://github.com/Yhe9718/GoLoCoNet .

Objectives: Dynamic MRI of the upper vocal tract is increasingly used to study speech. Image segmentation is often required to analyse the organs of speech; however, manual segmentation is labour intensive and time consuming and automatic methods are being developed. In this paper, a new hybrid transformer network is proposed for such task. Methods: We introduce a deep learning-based decoder model termed “Progressively Refinement Decoding (PRORED).” This model incorporates a directional field (DF) module designed to capture the contour details of features. The acquired contour information is leveraged to refine the boundaries both between and within classes. By integrating the DF module at different stages of the decoder, features are enhanced progressively, ensuring a more detailed and accurate segmentation. Results: Our model is evaluated using a publicly accessible speech MRI dataset and a cardiac dataset. The metrics employed are the Dice coefficient and the Hausdorff distance. Results indicate that our model attains an average Dice coefficient of 97.78% and a Hausdorff distance of 6.84 mm. Additionally, our network was able to identify closure patterns more efficiently than the baseline network and previously published work. In addition, the model was also evaluated on a cardiac dataset, and achieved 91.90% dice score. Conclusions: The proposed model leads to a more accurate segmentation of speech MRI data and in particular allows for a better velopharyngeal closure study. The proposed model was also evaluated on a cardiac dataset and achieved competitive performance, showing its strong generalizability. Advances in knowledge: First model that utilizes vision transformer and progressive refinement decoder to segment dynamic speech MRI.

Accurate medical image segmentation plays a vital role in assisting diagnosis with quantifiable visual evidence. Due to the complex structure and diverse patterns in medical images, it is crucial to capture both short and long-range pixel relations. While transformers are adept at modeling long-range spatial dependencies in images, they struggle with learning local pixel relationships. To address this, we propose a deep learning network named GoLoCo-Net incorporating a dual decoder structure. More specifically, one decoder entails a Contextual Attention Feature Enhancement (CAFE) module to enhance the features for a broader capture of local and global contexts, whereas the other uses a Global-Guide-Local Feature (GGLF) module that leverages high-level features to enrich low-level features with a global context. The proposed method is evaluated on two dynamic MRI datasets and one multi-organ CT dataset. Experimental results show that the model achieves state-of-the-art performance across all three datasets. The code is available:https://github.com/Yhe9718/GoLoCoNet.

Purpose: To evaluate the diagnostic performance of quantitative measures derived from [ ¹⁸ F]FDG PET/CT, [ ¹⁸ F]FEC PET/CT, and DW-MRI in the detection of lymph node metastases in endometrial and cervical cancer with comparison to standard visual PET analysis with histology as the reference standard. Methods: Subanalysis of quantitative data from the prospective multicentre MAPPING study. Nodal and tumour SUV _max from [ ¹⁸ F]FDG PET/CT and [ ¹⁸ F]FEC PET/CT and ADC _mean from DW-MRI were documented. Nodal-to-tumour ratios (NTR) and SUV _max -to-ADC _mean ratio (STAR) were calculated. Optimal cut-offs of quantitative measures were compared to visual assessment on a regional basis using histopathology as the reference standard. Results: Scans from 112 patients (36 cervical and 76 endometrial cancers; 340 nodal regions) were eligible for quantitative image analysis. Lower ADC _mean on DW-MRI was observed in metastatic nodes for cervical cancer but not for endometrial cancer. Quantitative measures were significantly higher in malignant than benign nodal regions on [ ¹⁸ F]FDG PET/CT and [ ¹⁸ F]FEC PET/CT in endometrial cancer. SUV _max cut-offs showed similar performance to visual assessment in the diagnosis of metastatic lymph nodes in endometrial cancer whilst ADC _mean cut-offs showed significantly lower specificity than visual assessment. Interobserver agreement was excellent for SUV _max measurements on both [ ¹⁸ F]FDG PET/CT and [ ¹⁸ F]FEC PET/CT, but poor for ADC _mean on DW-MRI. Conclusion: Quantitative measures from [ ¹⁸ F]FDG PET/CT, [ ¹⁸ F]FEC PET/CT, or DW-MRI did not outperform visual assessment in the detection of nodal metastases in endometrial cancer. Therefore, the implementation of these quantitative measures as standalone diagnostic tools in routine clinical practice is not recommended.

Objective: MRI is the preferred spinal imaging modality, but lacks metabolic information, while SPECT/CT provides functional information on osteoblastic activity. We aimed to determine if retrospective fusion of SPECT and MRI (SPECT/MRI) enhances disease site identification and reader confidence compared with SPECT/CT or MRI in chronic back pain refractory to medical management. Methods: One hundred and eighteen patients with chronic back pain not responding to medical management and investigated with MRI and SPECT/CT were selected. Two readers compared SPECT/MRI images to sequential reviews of SPECT/CT and MRI, MRI, and SPECT/CT alone. Potential pain generator sites and diagnostic confidence for each modality were recorded. New findings on SPECT/MRI were reviewed at a multidisciplinary meeting (MDM), and potential management changes were recorded. Results: Diagnostic confidence for SPECT/MRI (4.99) was higher than sequential SPECT/CT and MRI (3.8), MRI (3.6), or SPECT/CT (3). SPECT/MRI did not detect new disc disease, but found more sites (1%) in patients with known disc disease compared with other modalities. SPECT/MRI detected more patients and sites of endplate and facet disease compared with other modalities. It found 25% more patients and 49.5% more sites for endplate disease; 31.3% more patients and 46.7% more sites for facet disease compared with sequential review of SPECT/CT and MRI. The MDM review did not change the management for new disc disease sites. However, in 19 (25.6%) patients with new endplate disease, 45 (38.1%) patients with additional endplate disease, 20 (16.9%) patients with new facet disease, and 45 (38.1%) patients with additional facet disease, management changes were adopted. Conclusions: SPECT/MRI in refractory back pain detects additional disc, endplate, and facet disease sites and new patients compared with sequential SPECT/CT, MRI, or SPECT/CT alone, potentially leading to management changes.

The use of real-time magnetic resonance imaging (rt-MRI) of speech is increasing in clinical practice and speech science research. Analysis of such images often requires segmentation of articulators and the vocal tract, and the community is turning to deep-learning-based methods to perform this segmentation. While there are publicly available rt-MRI datasets of speech, these do not include ground-truth (GT) segmentations, a key requirement for the development of deep-learning-based segmentation methods. To begin to address this barrier, this work presents rt-MRI speech datasets of five healthy adult volunteers with corresponding GT segmentations and velopharyngeal closure patterns. The images were acquired using standard clinical MRI scanners, coils and sequences to facilitate acquisition of similar images in other centres. The datasets include manually created GT segmentations of six anatomical features including the tongue, soft palate and vocal tract. In addition, this work makes code and instructions to implement a current state-of-the-art deep-learning-based method to segment rt-MRI speech datasets publicly available, thus providing the community and others with a starting point for developing such methods.

Objectives: To develop a single-slice numerical phantom with known myocardial motion, at several temporal and in-plane spatial resolutions, for testing and comparison of Cardiovascular Magnetic Resonance (CMR) feature tracking (FT) software. Methods: The phantom was developed based on CMR acquisitions of one volunteer (acquired cine, tagging cine, T1 map, T2 map, proton density weighted image). The numerical MRI simulator JEMRIS was used, and the phantom was generated at several in-plane spatial resolutions (1.4 × 1.4 mm² to 3.0 × 3.0 mm²) and temporal resolutions (20 to 40 cardiac phases). Two feature tracking software packages were tested: Medical Image Tracking Toolbox (MITT) and two versions of cvi42 (v5.3.8 and v5.13.7). The effect of resolution on strain results was investigated with reference to ground-truth radial and circumferential strain. Results: Peak radial strain was consistently undermeasured more for cvi42 v5.13.7 than for v5.3.8. Increased pixel size produced a trend of increased difference from ground-truth peak strain, with the largest changes for cvi42 obtained using v5.13.7 between 1.4 × 1.4 mm² and 3.0 × 3.0 mm², at 9.17 percentage points (radial) and 8.42 percentage points (circumferential). Conclusions: The results corroborate the presence of intervendor differences in feature tracking results and show the magnitude of strain differences between software versions. Advances in knowledge: This study shows how temporal and in-plane spatial resolution can affect feature tracking with reference to the ground-truth strain of a numerical phantom. Results reaffirm the need for numerical phantom development for the validation and testing of FT software.

Purpose: Validation of quantitative MR measures for myelin imaging in the postmortem multiple sclerosis spinal cord.Methods: Four fixed spinal cord samples were imaged first with a 3T clinical MR scanner to identify areas of interest for scanning, and then with a 7T small bore scanner using a multicomponent‐driven equilibrium single‐pulse observation of T₁ and T₂ protocol to produce apparent proton density, T₁, T₂, myelin water, intracellular water, and free‐water fraction maps. After imaging, the cords were sectioned and stained with histological markers (hematoxylin and eosin, myelin basic protein, and neurofilament protein), which were quantitatively compared with the MR maps.Results: Excellent correspondence was found between high‐resolution MR parameter maps and histology, particularly for apparent proton density MRI and myelin basic protein staining.Conclusion: High‐resolution quantitative MRI of the spinal cord provides biologically meaningful measures, and could be beneficial to diagnose and track multiple sclerosis lesions in the spinal cord.

Purpose: Preoperative nodal staging is important for planning treatment in cervical cancer and endometrial cancer, but remains challenging. We compare nodal staging accuracy of 18F-ethyl-choline-(FEC)-PET/CT, 18F-fluoro-deoxy-glucose-(FDG)-PET/CT, and diffusion-weighted-MRI (DW-MRI) with conventional morphologic MRI. Experimental Design: A prospective, multicenter observational study of diagnostic accuracy for nodal metastases was undertaken in 5 gyne-oncology centers. FEC-PET/CT, FDG-PET/CT, and DW-MRI were compared with nodal size and morphology on MRI. Reference standard was strictly correlated nodal histology. Eligibility included operable cervical cancer stage ≥ 1B1 or endometrial cancer (grade 3 any stage with myometrial invasion or grade 1–2 stage ≥ II). Results: Among 162 consenting participants, 136 underwent study DW-MRI and FDG-PET/CT and 60 underwent FEC-PET/CT. In 118 patients, 267 nodal regions were strictly correlated at histology (nodal positivity rate, 25%). Sensitivity per patient (n = 118) for nodal size, morphology, DW-MRI, FDG- and FEC-PET/CT was 40%*, 53%, 53%, 63%*, and 67% for all cases (*, P = 0.016); 10%, 10%, 20%, 30%, and 25% in cervical cancer (n = 40); 65%, 75%, 70%, 80% and 88% in endometrial cancer (n = 78). FDG-PET/CT outperformed nodal size (P = 0.006) and size ratio (P = 0.04) for per-region sensitivity. False positive rates were all <10%. Conclusions: All imaging techniques had low sensitivity for detection of nodal metastases and cannot replace surgical nodal staging. The performance of FEC-PET/CT was not statistically different from other techniques that are more widely available. FDG-PET/CT had higher sensitivity than size in detecting nodal metastases. False positive rates were low across all methods. The low false positive rate demonstrated by FDG-PET/CT may be helpful in arbitration of challenging surgical planning decisions.