Treffer: A CNN-Driven Approach for Injury Type and Severity Detection with Hospital Recommendations for Emergency Response.

Traumatic injuries are a leading cause of emergency department visits and can rapidly progress to life-threatening conditions without prompt, accurate assessment. In many pre-hospital and resource-limited settings, first responders and clinicians lack immediate access to specialists or advanced diagnostic tools, resulting in inconsistent triage decisions and treatment delays. There is an urgent need for an automated, image-based solution that standardizes injury evaluation, enabling rapid, data-driven support for emergency response teams. To address this need, we developed a desktop application that integrates a custom convolutional neural network (CCNN) classifier with a color-segmentation-based severity detector. The system operates on standard computing hardware using open-source Python libraries--Tkinter for the graphical interface, OpenCV for image processing, Keras for deep learning, and scikit-learn for benchmarking traditional machine-learning models. Users load a dataset of injury photographs, automatically extract class labels (e.g., hand, head, leg), and cache processed image arrays to expedite future runs. The application supports training and comparing multiple models--Support Vector Machine, Decision Tree, Random Forest, and the proposed CCNN--on normalized, shuffled, and one-hot encoded image data. The CCNN consistently outperforms traditional models, achieving perfect classification on a held-out test set. For real-time inference, the system resizes and normalizes a selected test image, predicts the injury class via the trained CNN, and applies HSV color thresholding to detect red regions indicative of bleeding or bruising. Contour analysis classifies severity as "minor" or "major" based on region size, and the application overlays annotated bounding boxes on the image. A text-based recommendation engine then retrieves class-specific hospital advice, displayed alongside the annotated image and prediction metrics. By combining high-accuracy classification, interpretable severity detection, and actionable recommendations within an intuitive GUI, this solution fulfills the critical need for standardized, rapid injury assessment in diverse emergency settings, ultimately supporting more consistent triage decisions and improving patient outcomes. [ABSTRACT FROM AUTHOR]