Multi-signal　output　biosensor　technologies　based　on　optical　visualization　and　electrochemical　or　other　sophis-ticated　signal　transduction　are　flourishing.　However,　sensors　with　multiple　signal　outputs　still　exhibit　some　limitations,　such　as　the　additional　requirement　for　multiple　regression　equation　construction　and　control　of　results.　Herein,　we　developed　a　sensitive　cascade　of　colorimetric-photothermal　biosensor　models　for　prognostic　management　of　patients　with　myocardial　infarction　with　the　assistance　of　an　artificial　neural　network　(ANN)　normalization　process.　A　cascade　enzymatic　reaction　device　based　on　hollow　prussian　blue　nanoparticles　(h-PB　NPs),　and　a　portable　smartphone-adapted　signal　visualization　platform　were　integrated　into　the　all-in-one　3D　printed　assay　device.　Specifically,　liposomes　encapsulated　with　h-PB　were　confined　to　the　test　cell　using　a　classical　immunoassay.　Based　on　the　peroxidase-like　activity　of　h-PB,　the　h-PB　obtained　by　the　immunization　process　was　further　transferred　to　the　TMB-H2O2　system　and　used　as　a　cascade　of　signal　amplification　for　sen-sitive　determination　of　cTnI　protein.　The　target　concentration　was　converted　into　a　measurable　temperature　signal　readout　under　808　nm　NIR　laser　excitation,　and　the　absorbance　of　the　TMB　(ox-TMB)　system　at　650　nm　was　recorded　simultaneously　as　a　reference　during　this　process.　Interestingly,　a　parallel　3-layer,　64-neuron　ANN　learning　model　was　built　for　bimodal　signal　processing　and　regression.　Under　optimal　conditions,　the　bimodal　machine　learning-assisted　co-immunoassay　exhibited　an　ultra-wide　dynamic　range　of　0.02-20　ng　mL-1　and　a　detection　limit　of　10.8　pg　mL-1.　This　work　creatively　presents　a　theoretical　study　of　machine　learning-assisted　multimodal　biosensors,　providing　new　insights　for　the　development　of　ultrasensitive　non-enzymatic　biosensors.