Visual　quality　assessment　of　screen　content　images　(SCIs)　has　emerged　as　a　significant　research　field　in　digital　image　processing.　Current　methods　based　on　convolutional　neural　networks　(CNN)　and　Transformer　have　exhibited　promising　performance.　However,　most　of　the　existing　methods　only　extract　local　features　or　global　contextual　features　from　distorted　images　for　quality　perception,　lacking　the　information　interaction　between　local　and　global　features.　Furthermore,　due　to　the　limited　training　data,　most　methods　use　image　chunking　for　data　enhancement,　ignoring　the　visual　quality　differences　among　different　image　patches.　Aiming　at　the　above　problems,　our　paper　proposes　an　effective　no-reference　image　quality　assessment　model　for　SCIs　based　on　hybrid　CNN-transformer　and　multi-region　selective　features.　Firstly,　we　extract　the　local　and　global　contextual　features　of　multiple　regions　separately.　Then,　we　use　the　bilateral　feature　interaction　module　proposed　to　enhance　the　information　exchange　between　local　and　global　features,　thereby　refining　the　image　features　with　different　granularities.　Secondly,　to　fully　consider　the　visual　quality　differences　among　different　regional　image　patches,　we　propose　a　regional　feature　selection　module　to　adaptively　learn　the　correlation　between　them,　assigning　higher　learning　weights　to　important　regional　features.　The　experimental　results　show　that　the　model　proposed　in　this　paper　outperforms　the　state-of-the-art　no-reference　and　full-reference　SCI　quality　assessment　methods,　and　achieves　higher　consistency　with　the　subjective　visual　perception.　©　2023　IEEE.