Comprehensive Summary
While computed tomography is the major method used for renal cancer identification and assessment, it is largely dependent on manual interpretation, which makes it susceptible to human error. This process is further complicated by the fact that renal tumors exhibit variable appearance and indistinct margins. In this study, an attempt was made to automate the renal tumor sectioning process with the introduction of the DeepMedic 3D system. This system was trained with 5-fold cross validation of 382 contrast-enhanced CT scans which were manually annotated by physicians so that the system would emphasize particular areas of interest in the scans. The results show that this system exhibits high performance in kidney segmentation, with an average Dice coefficient of 93.82 ± 1.38%, precision of 94.86 ± 1.59%, and recall of 93.66 ± 1.77%. However, in renal tumor segmentation, the model attained a Dice coefficient of 88.19 ± 1.24%, precision of 90.36 ± 1.90%, and recall of 88.23 ± 2.02%. It is valuable to note that this study has no comparison to manual sectioning, so while these numbers seem high, there is no gauge to judge the impact on accuracy.
Outcomes and Implications
As we visit the results of this study and the lack of information relating to pure manual sectioning work, we are left with a potential ethical dilemma. Will the medical field prioritize a faster option to see more patients and reduce cost and overtime, or will it emphasize the quality and accuracy of diagnoses? In a similar vein, will the medical field allow the historical discrepancies in medical care quality for individuals of varying race and gender to penetrate this new era of medicine? It is vital that the training of this AI system prioritizes the utmost accuracy, especially for patients of varying race and gender. Additionally, it is more important now than ever that the medical field does not allow the thirst for automation to cut the radiologist out of the equation.