Since several decades, the physicians are able to access hollow organs with endoscopic methods, which serve both as diagnostic and surgical means in a wide range of disciplines of the modern medicine (e.g. urology, pneumology, gastroenterology). Unfortunately, white light (WL) endoscopy displays a limited sensitivity to early pre-cancerous lesions. Hence, several endoscopic methods based on fluorescence imaging have been developed to overcome this limitation. Both endogenous and exogenously-induced fluorescence have been investigated, leading to commercial products. Indeed, autofluorescence bronchoscopy, as well as porphyrin-based fluorescence cystoscopy, are now on the market. As a matter of fact, fluorescence-based endoscopic detection methods show very high sensitivity to pre-cancerous lesions, which are often overlooked in WL endoscopy, but they still lack specificity mainly due to the high false-positive rate. Although most of these false positives can easily be rejected under WL observation, tissue abnormalities such as inflammations, hyperplasia, and metaplasia are more difficult to identify, often resulting in supplementary biopsies. Therefore, the purpose of this thesis is to study novel, fast, and convenient method to characterize fluorescence positive spots in situ during fluorescence endoscopy and, more generally, to optimize the existing endoscopic setup. In this thesis, several clinical evaluations were conducted either in the tracheo-bronchial tree and the urinary bladder. In the urinary bladder, fluorescence imaging for detection of non-muscle invasive bladder cancer is based on the selective production and accumulation of fluorescing porphyrins, mainly protoporphyrin IX (PpIX), in cancerous tissues after the instillation of Hexvix® during one hour. In this thesis, we adapted a rigid cystoscope to perform high magnification (HM) cystoscopy in order to discriminate false from true fluorescence positive findings. Both white light and fluorescence modes are possible with the magnification cystoscope, allowing observation of the bladder wall with magnification ranging between 30× – for standard observation – and 650×. The optical zooming setup allows adjusting the magnification continuously in situ. In the high magnification regime, the smallest diameter of the field of view is 600 microns and the resolution is 2.5 microns, when in contact with the bladder wall. With this HM cystoscope, we characterized the superficial vascularization of the fluorescing sites in WL (370–700 nm) reflectance imaging in order to discriminate cancerous from non-cancerous tissues. This procedure allowed us to establish a classification based on observed vascular patterns. 72 patients subject to Hexvix® f luorescence cystoscopy were included in the study. Comparison of HM cystoscopy classification with histopathology results confirmed 32/33 (97%) cancerous biopsies, and rejected 17/20 (85%) non-cancerous lesions. No vascular alteration could be observed on the only positive lesion that was negative in HM mode, probably because this sarcomatoid carcinoma was not originating in the bladder mucosa. We established with this study that a magnification ranging between 80× and 100× is an optimal tradeoff to perform both macroscopic PDD and HM reflectance imaging. In order to make this approach more quantitative, different algorithms of image processing (vessel segmentation and skeletonisation, global information extraction) were also implemented in this thesis. In order to better visualize the vessels, we improved their contrast with respect to the background. Since hemoglobin is a very strong absorber, we targeted the two hemoglobin absorption peaks by placing appropriate bandpass filters (blue 405±50 nm, green 550±50 nm) in the light source. HM cystoscopy was then performed sequentially with WL, blue and green illumination. The two latter showed higher vessel-to-background contrast, identifying different layers of vascularization due to the light penetration depth. During fluorescence cystoscopy, we often observed that the images are somehow "blurred" by a greenish screen between endoscope tip and bladder mucosa. Since this effect is enhanced by the urine production, it is more visible with flexible scopes (lower flushing capabilities) and imaging systems that collect only autofluorescence as background. Indeed, when the bladder is not flushed regularly, greenish flows coming out of the ureters can easily be observed. For this reason, it is supposed that some fluorophores contained in the urine are excited by the photodetection excitation light, and appear greenish on the screen. This effect may impair the visualization of the bladder mucosa, and thus cancerous lesions, and lowers sensitivity of the fluorescence cystoscopy. In this thesis, we identified the main metabolites responsible for the liquid fluorescence, and optimized the spectral design accordingly. In the tracheo-bronchial tree, the fluorescence contrast is based on the sharp autofluorescence (AF) decrease on early cancerous lesions in the green spectral region (around 500 nm) and a relatively less important decrease in the red spectral region (> 600 nm) when excited with blue-violet light (around 410 nm). It has been shown over the last years, that this contrast may be attributed to a combined effect of epithelium thickening and higher concentration of hemoglobin in the tissues underneath the (pre-)cancerous lesions. In this thesis, we contributed to the definition of the input design of several new prototypes, that were subsequently tested in the clinical environment. We first showed that narrow-band excitation in the blue-violet could increase the tumor-to-normal spectral contrast in the green spectral region. Then, we quantified the intra- and inter-patient variations in the AF intensities in order to optimize the spectral response of the endoscopic fluorescence imaging system. For this purpose, we developed an endoscopic reference to be placed close to the bronchial mucosa during bronchoscopy. Finally, we evaluated a novel AF bronchoscope with blue-backscattered light on 144 patients. This new device showed increased sensitivity for pre-neoplastic lesions. Similar to what we observed in the bladder, it is likely that developing new imaging capabilities (including vascular imaging) will facilitate discriminating true from false positive in AF bronchoscopy. Here, we demonstrated that this magnification allowed us to resolve vessels with a diameter of about 30 µm. This resolution is likely to be sufficient to identify Shibuya's vascular criteria (loops, meshes, dotted vessels) on AF positive lesions. This criteria allow him to recognize pre-cancerous lesions, and thus can potentially decrease the false-positive rate with our AF imaging system. This magnification was also showed to be better for routine bronchoscopy, since it delivers sharper and more structured images to the operator.