Alexander Sieradzki

Background: Accurate vertebral instance segmentation on full-spine radiographs is essential for spinal parameter assessment, but supervised methods require costly instance-level annotations and may be sensitive to domain shift. Methods: We investigated whether promptable segmentation foundation models can generalize zero-shot to raw DICOM spine radiographs without task-specific training. We evaluated SAM-ViT-Huge, SAM2-Hiera-Large, and MedSAM-ViT-Base on 144 full-spine radiographs with 1309 annotated vertebral masks using a standardized pipeline for DICOM decoding, intensity normalization, automatic prompt generation, and instance-level evaluation. For each prompt, models produced three candidate masks. Performance was reported under an oracle protocol selecting the candidate with the highest IoU against ground truth and a model-score protocol selecting the candidate with the highest predicted IoU. Metrics included IoU, Dice, precision, recall, ASSD, and HD95. Results: The best configuration was SAM-ViT-Huge with rectangle prompting, reaching a mean IoU/Dice of 0.782/0.870 under oracle selection and 0.737/0.837 under model-score selection. SAM2-Hiera-Large with rectangle prompting followed (0.744/0.848 oracle; 0.699/0.815 model-score), ahead of MedSAM-ViT-Base (0.599/0.737 oracle; 0.387/0.499 model-score). Point prompting yielded consistently low overlap (IoU 0.224–0.319; Dice 0.276–0.414) despite high recall, indicating systematic over-segmentation and large boundary errors. Conclusions: Zero-shot vertebral instance segmentation on raw DICOM spine radiographs is feasible with promptable foundation models when prompts sufficiently constrain target extent. Rectangle prompting is clearly more effective than point prompting in this setting.

The furniture manufacturing sector faces significant challenges in machining composite materials, where quality issues such as delamination can lead to substandard products. This study aims to improve the classification of drilled holes in melamine-faced chipboard using Explainable AI (XAI) techniques to better understand and interpret Convolutional Neural Network (CNN) models’ decisions. We evaluated three CNN architectures (VGG16, VGG19, and ResNet101) pretrained on the ImageNet dataset and fine-tuned on our dataset of drilled holes. The data consisted of 8526 images, divided into three categories (Green, Yellow, Red) based on the drill’s condition. We used 5-fold cross-validation for model evaluation and applied LIME and Grad-CAM as XAI techniques to interpret the model decisions. The VGG19 model achieved the highest accuracy of 67.03% and the lowest critical error rate among the evaluated models. LIME and Grad-CAM provided complementary insights into the decision-making process of the model, emphasizing the significance of certain features and regions in the images that influenced the classifications. The integration of XAI techniques with CNN models significantly enhances the interpretability and reliability of automated systems for tool condition monitoring in the wood industry. The VGG19 model, combined with LIME and Grad-CAM, offers a robust solution for classifying drilled holes, ensuring better quality control in manufacturing processes.

This evaluation of deep learning and traditional machine learning methods for tool state recognition in milling processes aims to automate furniture manufacturing. It compares the performance of long short-term memory (LSTM) networks, support vector machines (SVMs), and boosting ensemble decision trees, utilizing sensor data from a CNC machining center. These methods focus on the challenges and importance of feature selection, data preprocessing, and the application of tailored machine learning models to specific industrial tasks. Results show that SVM, with an accuracy of 96%, excels in handling high-dimensional data and robust feature extraction. In contrast, LSTM, which is appropriate for sequential data, is constrained by limited training data and the absence of pre-trained networks. Boosting ensemble decision trees also demonstrate efficacy in reducing model bias and variance. Conclusively, selecting an optimal machine learning strategy is crucial, depending on task complexity and data characteristics, highlighting the need for further research into domain-specific models to improve performance in industrial settings.