Deep Learning-Based Hyperparameter Tuning and Performance Comparison

The accurate detection of small objects holds critical significance in many applications; however, challenges persist in this domain. In this study, we evaluated the effectiveness of deep learning algorithms used in object detection for medical image analysis. Additionally, we conducted Meningioma detection by examining the importance of hyperparameter tuning and its impact on performance using the YOLO deep learning model. While the majority of Meningiomas are benign and non-cancerous, they can induce various neurological symptoms by forming in proximity to brain tissues. For Meningioma detection, the YOLO model was tested using random search and grid search algorithms to identify critical hyperparameters with different parameter combinations. The findings demonstrated that implementing appropriate hyperparameters can significantly enhance small object detection tasks. A detection rate of 99.1% was achieved in Meningioma diagnosis. With accurate hyperparameter values, the model exhibited the capability to precisely identify small objects and substantially reduce false positives. Furthermore, at the conclusion of the study, we presented the recommended hyperparameter combinations for achieving the best performance. This study represents a critical step in designing adjustable and effective models in the field of deep learning-based small object detection. The obtained results underscore the potential of using YOLO in Meningioma detection and similar medical diagnosis applications, paving the way for developing more efficient and precise diagnostic methods in the future, while emphasizing the significance of hyperparameter tuning.