Minimal invasive procedures are often preferred over open surgeries because they result in smaller scars, fewer complications and a quicker recovery of the patients. However, they come at the cost of reduced mobility and perception limitations of the surgeon. In many laparoscopic surgeries, for example, it is necessary to stop the blood flow to a specific organ or tissue region by clamping the arteries responsible for blood supply. This procedure, commonly referred to as ischemia induction, prevents excessive bleeding of patients [Tho+07] and is performed in various procedures, including partial nephrectomy, organ transplantation and anastomosis. After clamping the main arteries, it is highly challenging to assess the perfusion state of the tissue solely based on the available RGB video stream. This holds especially true when selective clamping of a segmental artery is performed, in which ischemia is induced only in the cancerous part of the kidney during partial nephrectomy [McC+14; Bor+13] Traditional approaches to improving surgical vision involve fusing preoperatively acquired images with the situs [MH+18]. Such "offline" methods, however, cannot react on dynamics. The most common approach to ensure correct clamping is based on indocyanine green (ICG) fluorescence: after ICG is injected into the blood stream, it binds to the plasma. The bound ICG travels through the blood stream and accumulates in the internal organs, especially in the kidney and liver, within a minute [Tob+12; Gan+16].

Purpose: Optical imaging is evolving as a key technique for advanced sensing in the operating room. Recent research has shown that machine learning algorithms can be used to address the inverse problem of converting pixel-wise multispectral reflectance measurements to underlying tissue parameters, such as oxygenation. Assessment of the specific hardware used in conjunction with such algorithms, however, has not properly addressed the possibility that the problem may be ill-posed. Methods: We present a novel approach to the assessment of optical imaging modalities, which is sensitive to the different types of uncertainties that may occur when inferring tissue parameters. Based on the concept of invertible neural networks, our framework goes beyond point estimates and maps each multispectral measurement to a full posterior probability distribution which is capable of representing ambiguity in the solution via multiple modes. Performance metrics for a hardware setup can then be computed from the characteristics of the posteriors. Results: Application of the assessment framework to the specific use case of camera selection for physiological parameter estimation yields the following insights: (1) Estimation of tissue oxygenation from multispectral images is a well-posed problem, while (2) blood volume fraction may not be recovered without ambiguity. (3) In general, ambiguity may be reduced by increasing the number of spectral bands in the camera. Conclusion: Our method could help to optimize optical camera design in an application-specific manner.