Image sensors for scientific applications face unusually demanding performance requirements for speed, sensitivity, noise, dynamic range and data throughput. Dynamic proton radiography, for example, requires both extremely high frame rates (multi-MHz) and high dynamic range (~80 dB). Electron microscopy, particle physics, nuclear science and astrophysics applications requires high sensitivity and low noise for single-photon/single charged particle efficiency, often in extremely large (>100 M-pixel) arrays with tremendous sustained throughput (>50 G-pixels/s) and radiation-tolerance. This paper covers recent research results that extend standard CMOS image sensor performance in these areas. Signal to noise ratios as high as 90 dB have been achieved using a new, low-overhead, column-level active reset technique. This image sensor achieves output noise levels of ~45 microvolts, rms, without the use of correlated double sampling. Enhanced sensitivity for single-photon detection has been obtained by using an epitaxial silicon region as a higher cross-section CMOS sensor, with low-capacitance diode or large-area photogate charge collection. This development has made possible the use of standard 0.25 micrometer digital CMOS sensor arrays in place of expensive hybrid high-resistivity silicon sensor focal plane arrays plus CMOS readout circuit combinations. High-speed transient image sensors with frame rates as high as 10 M-frames/s, combined with 13-bit resolution, has been achieved using on-focal-plane frame storage. Fabricated in a standard digital 0.35 micrometer digital CMOS technology, the latter device includes a photodetector, a charge-integrating amplifier and an array of 64 sample circuits per pixel. Finally, a 100+ M-frames/s solid-state "streak camera" prototype is discussed which, in conjunction with the possible use of 3-D packaging techniques, may yield a sensor array capable of acquiring images at rates exceeding 10 T-pixels/s.