Detection of cancerous cells in the ovaries with deep İcaming algorithms

This study evaluates the performance of various deep learning architectures (ResNet-50, EfficientNetB0, DenseNet121, MobileNetV2, and InceptionV3) for classifying ovarian cancer cells. Each model was trained and tested using data from five different subtypes of ovarian cancer (Clear Cell, Endometrioid, Mucinous, Non_Cancerous, Serous). The dataset was divided into three main parts: training, validation, and testing, with data augmentation techniques employed to increase the diversity of the training data. During the training process, pretrained weights were loaded, the top layers were removed, and custom classifier layers were added. The performance of these methods was evaluated using class weight calculations, categorical cross-entropy loss, and fine-tuning techniques. The models' performance was illustrated with performance graphs, class-based results, and confusion matrices, and assessed using accuracy, precision, recall, and F1-score metrics. The results indicated that fine-tuned models generally performed better and that certain models were more suitable for specific classes. The EfficientNetB0 model achieved high accuracy rates across all classes. The overall accuracy of this model reached 98.87%, suggesting that the proposed method for cancer classification is promising. Following EfficientNetB0, the ResNet-50 model showed the second-highest performance. ResNet-50 achieved high precision, recall, F1 score, and accuracy, with macro-average values higher than those of the other models. The performance of the other models was inferior to these two models. These findings support the consideration of ResNet-50 and EfficientNetB0 as the recommended models. The results of the study demonstrate that deep learning models can be effectively used in the classification of ovarian cancer and can serve as a valuable guide for future research. These outcomes represent an important step towards developing more effective and precise diagnostic methods.