Clinical experts play a crucial role in screening and diagnosing squamous precancers as they integrate and interpret increasing amounts of multiscale and multimodal medical data. For example, detecting precancerous lesions may involve the use of widefield imaging modalities, with a large field of view and modest spatial resolution. In contrast, histologic diagnosis of precancer may involve modalities with high spatial resolution and small field of view, such as light microscopy. Additionally, newly available molecular tests can provide valuable patient-specific information that can help confirm a diagnosis or tailor a treatment to an individual patient. Due to the complexity and ever-growing amount of information available, efforts are underway to develop computer-aided diagnostic (CAD) systems to aid clinicians in the interpretation of medical data. However, most proposed CAD systems focus on applications involving a single modality or multiple modalities with similar spatial scales. The work presented in this thesis aimed to develop novel approaches to integrate multimodal, multiscale imaging and biomarker data for squamous precancer detection and diagnosis. Specifically, this thesis describes work to develop: (1) a deep learning model to diagnose high-grade squamous precancers using high-resolution endomicroscopy images, with performance validation across multiple anatomical sites, (2) a multiscale optical imaging fusion network that integrated high-resolution endomicroscopy images and widefield colposcopy data to diagnose cervical precancers, with validation in a large study in Brazil, and (3) a fusion and analysis framework integrating optical imaging data with DNA molecular test results and gene expression profiling for more accurate precancer diagnosis and prediction of the risk of lesion progression.