Summary This research work endeavor to suggest a new thyroid detection system that involves three main processes like segmentation, feature extraction, and detection. Initially, for the segmentation process, an improved watershed algorithm is proposed. Besides, from the segmented image, it extracted the texture features that comprises local binary patterns, modified GLCM feature, local tetra pattern features, together with the statistical features. Primarily, the extracted features are given as an input to convolutional neural network (CNN) for the final detection result. For making the detection more accurate, the training of deep CNN is performed via a new deer encircling included gray wolf optimization model for optimal weights tuning. The adopted hybrid algorithm combines gray wolf optimization and deer hunting optimization algorithm. At last, the outcomes of the adopted scheme is validated with the extant approaches under various metrics.

Summary In this article, the modified local binary patterns based feature extraction and hyper parameters tuned attention segmental recurrent neural network classifier with flamingo search optimization algorithm (MLBPFE‐ASRNNFSOAC‐DDM) is proposed for the disease diagnosis model. Here, breast cancer, diabetic, chronic kidney diseases diagnosis model is implemented. Initially, images are considered as dataset for disease diagnosis, which is given to the altered phase preserving dynamic range compression (APPDRC) scheme for preprocessing process. This APPDRC is used to preserve local features for boundary detection, thus; recovers image quality. Then the morphological, grayscale statistical and Haralick texture features are taken from preprocessed image with the help of modified local binary pattern process. The extracted features are given with attention segmented recurrent neural network (ASRNN) for classification. Then the weight parameters of ASRNN classifier are optimized by flamingo search optimization algorithm (FSOA), which increases the classification accuracy. This simulation process is accomplished at MATLAB platform. The proposed method, during Diabetic Diagnosis Model attains higher accuracy 16.4%, 21.45%, 30.38%, and 21.01% compared with the existing methods like FE‐ResNetV2DNNMSOC‐DDM and CMVHHO‐DKMLC‐DDM, CNN‐CAD‐DDM, and AD‐CFDRI‐SDL, respectively. During breast cancer Diagnosis Model attains higher accuracy 27.3%, 20.56%, 31.34%, and 25.13% compared with the existing methods like CMVHHO‐DKMLC‐BCM and SVMFE‐OPFPSOC‐BCM, MLPNN‐CNN‐BCM, and DCNN‐EL‐BCM, respectively.