Explaining Deep Learning Decisions Via Fuzzy Inference System on Medical Images

Explaining deep learning models has become crucial in various domains, including medical image processing. Understanding these models’ decision making processes is complicated as deep learning act like a black box when it comes on decision making. This can also help in advancements of medical imaging analysis, by allowing for more accurate diagnostic and treatment decisions and, ultimately, better patient care. In this study, we addressed the requirement for interpretability, by fusing the strength of CNNs with a Takagi-Sugeno-Kang (TSK) fuzzy inference system. We have utilized ResNet50, a transfer learning form of CNN, to classify a batch of 1000 images from dataset of chest cancer into four classes: ’Adenocarcinoma,’ ’Squamous Cell,’ ’Normal,’ and ’Large Cell’. The resolution of each image was varied, which was afterwards rectified in the preprocessing step, and this model attained 94.5 percent accuracy. We tried to improve the transparency and dependability of the model’s decision-making process. In this regard, we applied PCA to minimize the dimensions of predicted probabilities obtained from CNN for looking insight into the CNN model’s decision making process. Fuzzy c-means clustering is used to group the decision depending on the number of fuzzy rules. We extracted important insights from CNN’s predictions and offer a thorough justification of its decision-making process by incorporating the TSK fuzzy inference system.