DFS: Distributed Fusion Segmentation for Kidney Tumor using CT scan and Clinical Data Modalities
Keywords:
Distributed Data Parallel; Kidney Tumour; Light Weight segmentation; GRU; Dense, Depth wise Separable, Stocastic Gradiant Descent, Distributed Fusion segmentation; CT scan ; clinical records;).Abstract
Kidney tumors are a prevalent form of cancer that affects 75,000 individuals annually. Segmentation of kidney tumors is a critical step in accurate diagnosis, treatment planning, and evaluation of treatment outcomes. However, due to the complexity of the kidney structure and the variability of tumor shapes and sizes, segmentation remains a challenging task.
(1) Background:
With traditional machine learning algorithms require feature engineering, deep learning techniques such as CNNs can automatically learn features from the CT scan. However, training these models 6requires a large number of parameters, which can be time-consuming, and the inference time can also
be high. Furthermore, CNN-based methods may struggle with small and irregularly shaped .
(2) Method: Our approach, called DFS (Distributed Fusion Segmentation), combines both CT scan and clinical data modalities in a distributed fusion model. The DFS model consists of two main modules: the CT-based segmentation module and the clinical data-based segmentation module.
The CT-based module uses a deep convolutional neural network (CNN) to segment kidney tumor regions from CT scans, while the clinical data- based module uses various clinical data modalities, including patient age, gender, tumor location, and histology, to classify the kidney risk during its progression. The two modules then undergo a distributed fusion process that combines the segmented regions from both modules to produce a final segmentation mask.
(3) Results: The DFS model was trained and tested on a large dataset of CT scans and clinical data from kidney tumor patients, and it achieved state-of-the-art performance in tumor segmentation.
(4) Conclusion: Our results demonstrate that the integration of clinical data modalities into the segmentation process can significantly improve segmentation accuracy and reduce false positives. Moreover, the DFS model’s distributed fusion approach allows for a more comprehensive and accurate segmentation of kidney tumours, which can lead to improved diagnosis, 22treatment planning, and evaluation of treatment outcomes.
