Nanotechnology has been rapidly developing for early diagnosis and treatment of cancer, with nanoparticles being a large focus. However, traditional drug delivery mechanisms are passive and inefficient, with only 0.7% of nanoparticles reaching the tumour through blood vasculature. In vivo computation, also known as computational nanobiosensing (CONA), replaces nanoparticles with swarms of externally manipulable nanorobots whose movement is controlled by an external actuating system. The biological problem of smart tumour targeting is viewed from the computational perspective as an optimisation problem: nanorobot swarms (computational agents) explore the blood vasculature of high-risk tissue (search space) to locate the tumour (global optimum). Tumour biological gradient fields (BGF) create a fitness landscape, which can be analysed with fitness landscape analysis (FLA) to select and tune appropriate search algorithms for in vivo computation. Key limitations of previous work for in vivo computation are a lack of realistic BGFs that reflect the tumour microenvironment to test search algorithms on; and for FLA, no available measures that consider physical constraints of the in vivo environment. Two realistic tumour BGF models were created using COMSOL Multiphysics software (CFD Module), one highly vascularised, the other less vascularised. The vascular architecture was based on in vivo blood vessel networks in healthy and tumour regions, and blood velocity was used as a BGF. Blood velocity was found to be lowest in the tumour region, not exceeding 100 µm/s, confirming its applicability as a BGF for search algorithm testing. Three new FLA measures were created and validated with numerical simulations on two possible tumour vascular landscapes. These measures addressed the physical constraints of discrete search space, unidirectional blood flow, and nanorobot steering imperfections when using a uniform magnetic field. The less vascularized landscape was found to be more discrete, more heterogeneous, and contain a smaller countercurrent frequency of search direction. This indicated it would be more challenging to solve for than the highly vascularised landscape. These advancements of the CONA framework allow the in vivo search environment to be better visualised and understood for algorithmic development, as well as provide realistic BGFs to test these search algorithms on.