<p>Dialogue summarization becomes a difficult task due to a variety of factors. First, the majority of the pertinent data in a discussion is dispersed across statements due to multi-party exchanges with various textual patterns. Secondly, the dialogues are commonly informal in which diverse persons convey personal opinions, compared to a text summarization, which typically targets official publications like newspapers. Hence to understand and to summarize the medical-related dialogues from the telehealth system, a new deep learning-based dialogue summarization framework is introduced in this paper. At first, the required input data are gathered from the standard websites. The attained input is then pre-processed for removing noises and unwanted characters from the gathered data in order to reduce the complications that arise in the summarization process. The pre-processed data are further provided to the adaptive trans-residual bidirectional long short-term memory (Bi-LSTM) (ATR-Bi-LSTM) unit with encoder–decoder architecture to generate the final summarized outcome. The parameters in the ATR-Bi-LSTM are tuned with the aid of the improved Young’s double-slit experiment optimizer (IYDSEO). The final summarized dialogues are obtained from the developed ATR-Bi-LSTM model. The experimental verification is carried out in order to prove the effectiveness of the implemented deep learning-based dialogue summarization model.</p>